We measured PM₄ crystalline silica concentrations following NIOSH Method 7500 (NIOSH, 1994) outside of 17 residential homes within 800 m of the property line of facilities with active sand mining, processing, and/or transport in Trempealeau County, WI. These homes were selected randomly from a list of volunteering owners solicited through mail. Eligible homes were identified through publicly-available county property records. Sampling was conducted outside the home in a location mutually agreed upon by the investigator and the resident with the goal to sample as far away from trees and structures (e.g., shed, house) as feasible. Air was drawn with pumps (Omni 400, BGI by MesaLabs, Butler, NJ) at 4.2 L/min through PM₄ samplers (GK2.69 cyclone, BGI by MesaLabs, Butler, NJ) held 1.5-m from the ground and fitted with PVC filters (37-mm, 5.0-μm pore, Cat No 225-5-37, SKC Inc., Eighty-Eight, PA) for a minimum of 48 h. Sampling at this height with a respirable sampler represented the potential exposure of a resident when outdoors. Wind speed and direction were monitored with equipment mounted on the same pole as the air sampler. Samples were analyzed first gravimetrically (NIOSH Method 0600) (NIOSH, 1994) and then for crystalline silica by XRD (NIOSH Method 7500) (NIOSH, 1994). These sampling and analysis methods were selected to provide a minimum reporting limit (defined as five times the minimum detectable level) for crystalline silica concentration of 0.4 μg/m³.

Six sites were selected for real-time PM monitoring using the same criteria specified for crystalline silica air sampling above. Detailed information on these sites and sampling methodology are provided in Supplemental information, Section B. Briefly, three sites were near mines with onsite processing, two were near train trans-load/transportation facilities, and one was near a processing-only facility (Supplemental Information, SI, Table S1). At each site the monitoring equipment was set up as far from trees as possible and with a clear sightline to the nearest facility. The sampling time at each site averaged 17 days (range: 7–26). The overall sampling time frame (June 2014 – November 2014) aligned with the mining season. During the sampling time frame, typical summer and early fall weather conditions occurred.

An optical particle counter (OPC, Environmental Dust Monitor, Model EDM164, GRIMM Aerosols Technik, Ainring, Germany), positioned inside a trailer (SI, Figure S1), was used to measure PM₁₀ and PM_2.5. We selected this sampler because it provides highly temporally resolved PM concentrations for multiple size fractions. When relative humidity is greater than 50%, this sampler heats the sampled air to avoid positive bias from hygroscopic aerosols and to evaporate water droplets. A sampling head mounted 15 cm above the roof of the trailer was used to convey ambient aerosol to the OPC. External sensors mounted on a pole and attached to the side of the trailer were used to measure wind speed, wind direction, sound level, and motion. Wind speed and wind direction were measured with an ultrasonic anemometer (Model 85000, R.M. Young Company, Traverse, MI) mounted 3.7 m from the ground. A microcontroller was used to collect PM₁₀ and PM_2.5 from the OPC every 6 s and external sensor data every second. These measurements were averaged and logged to an SD card every 20 s.

MATLAB (8.3.0.532 R2014a, MathWorks, Natick, MA) was used to calculate PM_10-2.5 (calculated by subtracting PM_2.5 from PM₁₀), categorize explanatory variables, and remove unreadable records due to short power outages at Site 3 for less than 30 min. For each site, mean daily (from 12:00 am to 11:59 pm) concentrations were calculated as the arithmetic mean of 20 s. data. Statistical software (Minitab 17, Minitab Inc., State College, PA) was used to generate probability plots of 20-s PM_2.5 and PM_10-2.5 data. A bimodal distribution of concentrations was observed on probability plots for all sites (SI, Figure S2 for PM_10-2.5 and SI, Figure S3 for PM_2.5). Thus, a breakpoint concentration, defined as the maximum curvature (represented with an arrow in SI, Figure S2A), was determined to separate the two modes. Conditional probability plots of PM data versus wind direction data were then generated for PM data above the breakpoint concentration and for values above the NAAQS following Kim and Hopke.

The local contribution to PM_2.5 and PM_10-2.5 was estimated for 5-min, 1-hr, workday, and 24-hr periods following Watson and Chow (Watson and Chow, 2001). Local PM was calculated as the concentration observed for that time period subtracted by a successive moving average.

Raw sand (unprocessed sand from a mine) was collected from a site associated with the Upper Cambrian Jordan sandstone formation in Wisconsin known in the hydraulic fracturing industry as “Northern.” White. Proppant sands of four different mesh sizes (20/40, 30/50, 40/70, 100M) were also obtained from a proppant sand distributor.

To aerosolize the raw and proppant sands, 20.0 g of sand was weighed on a 4-place balance and placed in an acoustical dry aerosol generator/elutriator (ADAGE) system (Thorne, 1994). Dried, filtered, and compressed air flowed through the ADAGE system at 2.5 L/min into an enclosed chamber. Five samples of respirable dust from raw sand and each proppant type were collected with a respirable aluminum cyclone (SKC Inc., Eighty Four, PA) operating at 2.5 L/min for 60 min in accordance with the NIOSH Manual of Analytical Methods, Method 7500. The concentrations in the chamber were elevated to a level that would be expected to result in the majority of samples collected to exceed the limit of detection of Method 7500 (10 μg silica/filter sample) within the 60-min sampling period.

A total of 25 samples were collected on 5-μm-pore polyvinyl chloride (PVC) filters at 2.5 L/min using an SKC AirChek sampling pump (Model 22PCXR4, S/N 634006, SKC Inc, Eighty Four, PA). The sampling pump flow rate was calibrated before and after the sample collection period with a Gilibrator calibrator (S/N 1103045, Sensidyne, St Petersburg, FL. Filters from samples collected with the respirable cyclone were weighed before and after sample collection on a microbalance (Mettler Toledo XP-26, S/N 112122649, Mettler-Toledo LLC, Columbus, OH) following NIOSH Method 0600 (NIOSH, 1994) and then for crystalline silica by by x-ray diffraction (XRD) following NIOSH Method 7500 (NIOSH, 1994). Five g each of bulk, sieved raw sand (sieve sizes: 18/40, 40/60, 60/100, >100) and 5 g of each proppant type (sieve sizes: 20/40, 30/50, 40/70, 100M) were also analyzed by XRD for silica content. Additional cyclone filter samples were analyzed by scanning electron microscopy (SEM) with energy dispersive x-ray spectroscopy (EDS).

For purposes of visualizing the degree to which particulates may emenate from facilities associated with the frac sand industry in Wisconsin, air dispersion modeling was conducted for two sites: a processing plant and an open pit mine. The final evaluation of dispersion model results includes PM₁₀ predicted by the model in addition to the background level of PM₁₀. A background level of PM₁₀ for the state of Wisconsin of 29.4 μg/m³ was used in this analysis (WDNR, 2014). Dispersion modeling was performed with the EPA-approved dispersion model, AERMOD, incorporated in a vendor-supplied software package (AERMOD View^™ ver. 9.1.0, Lakes Environmental, Waterloo, Ontario). Modeling guidelines developed by the WDNR were followed (WDNR, 2015). Meteorological data, supplied by the WDNR (WDNR, 2016a), consisted of five years (2006 – 2010) of one-hour sets for sources within a region around the collecting meteorological station (Eau Claire, WI meteorological station). Upper-air meteorological data was measured at the St. Cloud Muncipal Airport.

The processing plant is located in Chippewa Falls, WI and represents a large facility dedicated to frac sand processing and shipping. In 2013, the plant consisted of sand dryers, product silos, a rail loadout area, truck receiving station, conveyers and stackers, material stockpiles and truck traffic on haul roads. Emissions for each of these potential sources of PM₁₀ were obtained from a 2012 WDNR air emissions inventory summary report for the plant (WDNR, 2012a). Locations for the sources were determined from aerial photos. The photos did not show a plant fenceline, therefore the plant boundary was estimated from the plant structures and roadways, which represented the area within which the public would not be expected to be allowed access. A variety of emission source types were used when applying these PM₁₀ sources to the dispersion model including: point (dryers, loadout and recieiving), area (conveyer and stackers), volume (product silos), and line (roads). At the time of the emissions inventory the stockpiles were uncovered and therefore represented a PM source from windblown dust. The two sand dryers utilized natural gas boilers which were not a source of PM but each dryer conveyed dried sand that produced PM. Baghouse filters were employed to control emissions during the conveying process as well as other point sources of dust emissions associated with the dryer building; the WDNR report therefore provided controlled PM emission rates from those sources. The report also mentions that watering trucks were used to suppress dust emissions from haul roads. All emission levels were reported for work schedule consisting of 24 hr/day, 7 days/wk, 365 days/yr.

The open pit mine is located in New Auburn, WI and was modeled as depicted from aerial photos obtained during 2014. PM emission sources within the mine consisted of a rock crusher, dryer, sand piles, open pits, and truck traffic on paved and unpaved haul roads. Emission rates for each of these potential sources of PM₁₀ were obtained from a 2013 WDNR air emissions inventory summary report for the mine (WDNR, 2013). Control technologies to suppress PM emissions at this mine were not listed in the WDNR summary report. The USEPA and state air quality bureaus offer less guidance for modeling open pit mines and sand piles compared to the information they make available for modeling typical industrial sources. Therefore, additional guidance was obtained from a modeling document written for the sand and gravel industry to provide the most appropriate modeling options for those source types (Heinerikson et al., 2007). For example, this information was used to establish AERMOD source types for processes utilized in sand mines such as crushers (area source), sand piles (area source) and open pits (open pit source).

A receptor grid was established according to WDNR guidance (WDNR 2016b). The terrain option was utilized with terrain data obtained from the Shuttle Radar Topography Mission (STRM) available through the US Geological Survey (USGS, 2016). Given five years of meteorological data, the WDNR guidance is to model the 6^th highest value for every grid node to mimic the standard requirement of no more than one excursion above the 150 μg/m³ limit per year on average over 3 years.

We analyzed publically-available PM_2.5 and PM₁₀ concentrations measured with federal reference or equivalent method samplers near or within the fenceline of facilities with sand mining activities. These measurements were either required or requested of the facility at the time of permit application by state agencies in Wisconsin and Minnesota. PM₁₀ was measured at 16 monitoring stations from Nov 2010 to Sep 2015 (overall n = 3,395; WI n = 2,549; MN n = 846). Total suspended particulate (TSP; n = 286 from 2 locations with one location having monitors north and south of a mine), PM_2.5 (n = 444 from 2 locations both having monitors north and south of a mine), and PM₄ crystalline silica (n = 417 from 4 locations with 2 locations having monitors north and south of a mine) were measured as part of sampling required by the Minnesota Pollution Control Agency from Jul 2012 to Oct 2015.

Respirable sampling was conducted over a 48-h at 17 residences near sand mines. The mean PM₄ was 9.1 μg/m³ (St Dev = 2.6 μg/m³), ranging from 6.0 μg/m³ to 15 μg/m³. Crystalline silica (α-quartz) was detected above the method limit of detection on 7 of the 17 samples. Of those samples, quartz represented 2% to 4% of the mass. All PM₄ crystalline silica concentrations were less than the minimum reporting level of 0.4 μg/m³, which is lower than the value of concern for chronic exposures adopted by CA and MN (3 μg/m³).

Long-term air monitoring at homes near sand mining and processing operations revealed PM concentrations well below the NAAQS for PM_2.5 and PM₁₀ (Table 1). Mean concentrations observed throughout monitoring ranged from 6.5 μg/m³ (Site 6) to 10 μg/m³ (Site 4) for PM_2.5 and from 11.1 μg/m³ (Site 3) to 19.8 μg/m³ (Site 5) for PM₁₀. The highest 24-h mean concentrations were fairly similar between sites for PM_2.5 (range: 11.4 μg/m³ at Site 1 to 14.5 μg/m³ at Site 2), whereas the range was larger for PM₁₀ (range: 18.6 μg/m³ at Site 1 to 38.0 μg/m³ at Site 5).

Breakpoint concentrations in the probability plots for 20-s data ranged from 18 μg/m³ to 36 μg/m³ for PM_2.5 and 22 μg/m³ to 45 μg/m³ for PM_10-2.5 (Table S2). Most of the concentrations (99% or more) were below these levels. In some cases, concentraions above the breakpoint (i.e., peak concentrations) were related to mining activities when using wind data (Figure 1 for PM_10-2.5 and SI, Figure S4 for PM_2.5). For example at Site 1 (Figure 1A), PM_10-2.5 was greater than the breakpoint concentration of 27 μg/m³ when winds were blowing over the mine. At Site 4 (Figure 1D), PM_10-2.5 was also greater than the breakpoint concentration (23 μg/m³) most often when the wind was blowing over the mine but not other operations (processing or conveying) or from directions without mining activity.

Maximum 24-hr local concentrations ranged from 0.5 μg/m³ to 4.1 μg/m³ for PM_2.5 and 0.2 to 18 μg/m³ for PM_10-2.5 (Table 2). These local contributions are low compared to NAAQS for PM_2.5 (35 μg/m³) and also for coarse particles, which are regulated under the NAAQS for PM₁₀ (150 μg/m³). As expected, the maximum local concentration increased as the averaging time was shortened, highlighting the fact that high concentrations do occur but over short duration.

PM₄ concentrations produced by aerosolizing raw sand were 18 times higher than concentrations produced by proppant sand (Table 3). In contrast, the percentage of crystalline silica in PM₄ produced by proppant sand (mean = 19.0%) was 3.3 times higher than that in raw sand (mean = 5.7%; Table 3).

Photographs of the bulk raw and proppant sand and sand grain sizing results from the sieve analysis are provided in SI Figure S5. Silica content in bulk samples were similar for raw and proppant sand types, ranging between 59 – 88%. Therefore, given the results shown in Table 3, the PM₄ size fraction of the bulk sand contained a much lower proportion of silica than non-respirable particles. The individual particle elemental composition of the PM₄ particles collected after aerosolization in the chamber revealed that there were PM₄ particles in both raw sand and proppant containing a high percentage of silicon and oxygen indicative of silicon dioxide, and therefore were assumed to be crystalline silica (representative spectra shown in Supplemental Information; SI, Figure S6). However, the large majority of PM₄ particles displayed an elemental profile of the mineral feldspar, which contains silicon, oxygen, aluminum and potassium (SI, Figure S6A). It can be conjectured that feldspar (hardness = 6) was crushed into particles in the PM₄ size fraction more easily than silicon dioxide (harness = 7) during the formation of the sandstone rock layer and therefore contributed a much larger percentage of all PM₄ particles.

PM10 outside of the the boundaries of the processing plant determined with dispersion modeling were below the NAAQS (Figure 2A). In contrast, the dispersion model for the sand mine indicated that this site was capable of exceeding the NAAQS for PM₁₀ within short distances beyond property lines. The 6^th highest PM₁₀ concentrations exceeded the NAAQS 40 meters from the mine (Figure 2B)..

Median PM concentrations measured by state agencies near facilities with sand mining activities were substantially below 24-hr regulatory standards or guideline values (Figure 3). No total suspended particulate measurements were above regulatory values (Figure 3A; Minnesota TSP standard of 260 μg/m³), and only 3 of 286 were higher than 150 μg/m³. Similarly, PM_2.5 was below the NAAQS of 35 μg/m³ (Figure 3C; N = 444). For PM₁₀ (Figure 3B), 2 (1 in WI and 1 in MN) of 3,395 measurements exceeded the PM₁₀ NAAQS of 150 μg/m³. All but 7 of 212 24-hr PM₄ crystalline silica mass concentrations observed at the Titan facility were higher than 3 μg/m³, the CA and MN guideline value for a one-year averaging period (Figure 3D). The MN EQB showed that these 7 concentrations occurred when winds were blowing over the site toward the samplers (MPCA, 2015).