2.1.

Site and Operation Description

This CMP $\eta$ study was performed during the spring tillage sequence following the harvest of a winter wheat crop in preparation for the planting of corn. It was carried out in the San Joaquin Valley of California during May and June 2008. Two adjacent fields were used with identical crop and flood irrigation treatments over the previous several years. Both fields were cultivated in winter wheat in late 2007 and were to be planted in corn for the 2008 summer growing season. The wheat was harvested while still green for silage 4 days before the tillage processes began, resulting in standing stubble but little plant material left on the surface. The site was chosen based on producer cooperation, historically dominant northwest winds, and field layout.

The surrounding landscape was topographically flat and dominated by agricultural production, including grain and corn fields, almond orchards, grape vineyards, and commercial dairy operations. Both fields were surrounded on all sides by roads. These roads, with the exception of one, were unpaved roads used for field access by farm machinery. The paved road, which was downwind of the fields during all measurements, was heavily traveled. Railroad tracks were located to the north of this site, with two to three trains passing by per day with varying numbers of cars. USDA NRCS soil survey data list the soil in both fields as soil type 130 – Kimberlina fine sandy loam, saline-alkali.¹⁶

The CMP selected for this study was the conservation tillage method. As described in Ref. 4, this CMP “involves using a system in which the soil is being tilled or cultivated to a lesser extent compared to a conventional system” and it is “intended to reduce primary soil disturbance operation such as plowing, disking, ripping, and chiseling.” The conservation tillage CMP under study was a strip–till method which combines multiple operations to reduce the number of passes required and only disturbs the soil in strips 0.2 m wide centered every 0.8 m instead of disturbing the entire surface. Strip–till reduced both the number of passes and the tilled surface by about 75%, as well as left most of the wheat stubble still standing for ground cover. The strip–till implement used in this study was the Orthman 1-tRIPr. The cooperating farm had been using the Orthman 1-tRIPr for seedbed preparation on all of its fields for several years, with the exception of field 4, which was prepared by conventional tillage methods.

The conservation tillage CMP applied in this study consisted of three tillage activities totaling three operations across the field, excluding the building and removal of ditches. All three were monitored in separate sample periods. In comparison, the conventional tillage method as applied here had nine different tillage activities totaling 13 operations, excluding the building and removal of a ditch and field-edge borders. Monitoring of 12 of the operations occurred over nine sample periods. Note that not all operations were active throughout sample periods in which multiple operations occurred. Also, not all of the conventional tillage operations are intended to work the entire field, such as breaking down in-field borders. The term in-field borders as used here applies to low ridges of soil that separate the field into smaller areas for flood irrigation. The conventional tillage method was employed in field 4, shown in Fig. 1, and the conservation tillage CMP was used in field 5. The operations that were performed in each management practice are shown in order in Table 1, with their corresponding dates, equipment utilized, number of passes over a given area, tractor run time, total area worked, and sample period length. In cases where multiple tractors and implements were used within a sample period, they are listed in the order of use with the area worked and tractor time being summed. The lister, also called a double plow, prepares the soil for planting by creating furrows and ridges. The cultivator passes in the conventional tillage sequence function as mechanical weed control, whereas a chemical weed control (herbicide) is used in the CMP sequence.

Fig. 1

Map of fields under study and the sample layout for each field. L1 and L2 represent light detection and ranging (LIDAR) locations used during sampling and dashed lines emanating from L1 and L2 show positions of vertical LIDAR scans.

Table 1

Information for each sample period regarding tillage operations, equipment used, tractor operation time, area worked and sample time.

During the first part of the lister operation, plant material not harvested along in-field borders caused clogging of the lister, decreasing effectiveness. A second tractor with a disk set was brought in to repeat the effort along the border lines to further reduce the size of residual plant material. Also note that the cultivator passes 1 and 2 and the roller pass for the first sample period on June 5 were not finished when planting in field 4 began and the second sample period was started. The combination of the cultivator and roller positions and meteorological conditions prevented significant impacts from these operations on downwind samples located near the southern end of the field.

Additionally, cultivator pass 4 was carried out the day after cultivator pass 3, but the emissions were not measured due to scheduling conflicts. It is assumed, in calculating the total PM emissions, that the ERs of both passes 3 and 4 were equal. In general, two cultivator passes are performed in sequence in opposite directions down the rows to ensure adequate weed control.

Ditches and field-edge borders were built and then broken down in both fields between May 20 and June 5 to allow for flood irrigation prior to planting. Water for irrigation was taken from the earthen holding pond of the adjacent dairy; drainage ditches on the east side of both fields returned excess water to the same holding pond. As the ditch and field-edge border construction and removal were not measured in the CMP field, the corresponding step for the conventional tillage method was not considered in the total emissions per method. Prior to any spring tillage activities, both fields 4 and 5 had in-field borders running in roughly an east/west direction. The in-field borders in field 5 were not broken down and smoothed out, but instead were used for the summer corn crop. However, in field 4 they were removed and the irrigation water moved in the furrows created by the lister.

Field personnel observed operations continually and recorded notes on tractor operation times, potential contamination issues due to traffic on surrounding dirt roads and wind-blown dust, general meteorological observations, and so on.

2.2.

Instrumentation and Sample Layout

A very dominant northwest wind was found in historical data for the months of May and June from a representative meteorological monitoring station in Stratford, California, in the California Irrigation Management Information System (CIMIS). Therefore, the PM and meteorology sampling layouts were configured to measure upwind conditions to the north and west and downwind conditions to the south and east.

Meteorological measurements were made at upwind and downwind locations with the instrumentation in Table 2. Vertical temperature, humidity, and wind speed profiles were measured using two 15.3 m towers, one upwind and one downwind as shown in Fig. 1. Each tower had five humidity/temperature sensors at 1.5, 2.5, 3.9, 6.2, and 9.7 m agl and 3-cup anemometers at 2.5, 3.9, 6.2, 9.7, and 15.3 m agl. Wind direction was measured at 15.3 m using a wind vane instead of the typical 10 m due to the fact that LIDAR measurements were made at higher elevations (up to 200 m agl) and the 15.3 m measurement height was reasoned to provide a better representation of both ground level and higher elevation wind direction than the 10-m height. Additionally, a meteorological station monitored wind speed, wind direction, temperature, relative humidity, precipitation, barometric pressure, and incoming solar radiation at 5 m agl at the air-quality trailer (AQT) location. Three pairs of three-dimensional sonic anemometers and infrared gas analyzers were deployed, one at an upwind location and one each downwind of the two fields of interest, to characterize upwind and downwind turbulences, as well as vertical fluxes of latent heat (evaporation), sensible heat, carbon dioxide, and horizontal momentum. Bulk density and soil moisture were quantified several times throughout the study, with calculations performed as described in Ref. 17.

Table 2

Manufacturer, precision, and accuracy information for deployed meteorological instrumentation.

PM mass concentrations were monitored by 20 MiniVol Portable Air Samplers (Models 4.2 and 5.0, AirMetrics, Eugene, Oregon), referred to hereafter as MiniVols. They are a portable, programmable, filter-based sampler that is battery-powered and yields an integrated sample over the exposure period. Filters were exposed for the duration of each sample period (see Table 1), yielding a single mass concentration measurement per sampler per sample period. Cumulative samples of particles up to ${\mathrm{PM}}_{2.5}$ or ${\mathrm{PM}}_{10}$ are accomplished via an impactor plate assembly inserted just upstream of the filter; total suspended particulate-matter (TSP) may be sampled without an impactor assembly in place. The MiniVol impactor assembly is designed to operate a flow of $5.0\mathrm{L}·{\min }^{-1}$, though the flow rate is neither actively monitored nor actively controlled by the system. It is set prior to deployment by the user via a calibrated rotameter. Rotameter flow calibration was performed prior to deployment. While several studies have found that ${\mathrm{PM}}_{2.5}$ and ${\mathrm{PM}}_{10}$ levels reported by MiniVols are very similar to concentrations measured by federal reference method (FRM, see 40 CFR 50.6 and 50.7) monitors, the slope of the particle removal efficiency versus particle size curve of the MiniVol impactor assembly is less steep than required by FRM samplers.^18–21 Therefore, PM levels reported by the MiniVols should be considered as close approximations to those that would be given by FRM samplers.

Pre- and postweights for the 47 mm Teflon filters used to collect particles were quantified using a calibrated microbalance (Type MT5, Mettler-Toledo, Inc., Columbus, Ohio). Filter conditioning was carried out in accordance with guidance in 40 CFR 50 Appendix J. Sample period average mass concentrations were calculated by dividing the total mass catch (average postweight minus average preweight) by the volume of air sampled.

Particle size distribution (PSD) was measured by eight Aerosol Profilers (Model 9722, Met One Instruments, Grants Pass, Oregon), also known as optical particle counters (OPCs). The OPC sums particle counts in eight size bins over nominal 20-s sample periods. The particle diameters ($d_p$) for lower bin limits were 0.3, 0.5, 0.6, 1.0, 2.0, 2.5, 5.0, and $10.0\mu \mathrm{m}$, with the last channel counting all particles $\ge 10.0\mu \mathrm{m}$. The factory calibrations of signal strength versus particle size using polystyrene latex beads of known size were used due to varying atmospheric aerosol composition. The sample flows were not conditioned prior to passing through the sampling chamber during these measurements due to the dry conditions, though this is suggested in atmospheres with high relative humidity. OPC flow measurements and inter-OPC count calibrations were performed onsite and applied in postanalysis. Particle volume concentrations ($V$) per bin were calculated from the counts, assuming spherical particles and using the geometric mean diameter (GMD) as the representative $d_p$. Values of $V$ in bins up to $d_p=k$ were summed to estimate the cumulative volume concentration ($V_k$).

The MiniVols and OPCs were deployed on towers and tripods upwind and downwind of the fields, as shown in Fig. 1. Most downwind sensors were moved between the downwind layouts, depending on the field being tilled. The AQT and the associated samplers did not move. The AQT is a $5\mathrm{m}\times 2.5\mathrm{m}\times 2.5\mathrm{m}$ cargo trailer used as the base of operations and equipped with tables, a refrigerator and dessicator for sample storage, and a rooftop platform for sensor deployment. Samplers were placed on tripods at 2 m agl at all locations except for those on top of the AQT at 5 m and those at the top of the towers at 9 m. Not all sample sites shown for a given setup were used in each sample period due to instrument availability limitations. However, samples were collected at a minimum of two upwind and six downwind locations during each sample period.

At most of the locations, multiple MiniVols with different impactor configurations and an OPC were collocated in order to characterize particle size and mass distributions. These data were used to calculate mass conversion factors (MCFs) for each size fraction ($k$), as described in detail by Ref. 21. In summary, the ${\mathrm{MCF}}_k$ is calculated using ${\mathrm{PM}}_k$ reported by MiniVols and $V_k$, averaged over the MiniVol sample time, from each sample location through the following equation:

The MCF is a simplified method to account for several complex and possibly interdependent variables that affect how an aerosol mixture is measured/detected based on both optical and aerodynamic properties. It incorporates many factors, such as particle shape, density, indices of refraction different from OPC calibration aerosols, and instrument sampling efficiencies, into a single scalar value. The MCF also includes effects due to optical systems measuring particles in ambient conditions while mass concentrations are calculated based on conditioned filters. This effect may be significant in humid environments, but the effect is assumed to be negligible in warm and dry conditions such as those found during this study. The ${\mathrm{MCF}}_k$ values were used to convert the OPC $V_k$ into ${\mathrm{PM}}_k$ to examine concentrations on a much finer temporal scale than possible with the filter data.

The Aglite LIDAR system was deployed to characterize PM concentrations in addition to point sensors. The Aglite LIDAR is a portable system using a micropulsed Nd:YAG laser with three wavelengths ($\lambda$), 355, 532, and 1064 nm. It has the capability to scan 280 deg in azimuth and from $-10\mathrm{deg}$ to $+45\mathrm{deg}$ in elevation. The effective range is 500 m to 15 km with each range bin approximately 6 m in length.²² The LIDAR was placed in crosswind positions 550 m from the nearest tillage area border. It was at L1, as shown in Fig. 1, from May 17 through June 11 and at L2 for the June 18 sample period. Critical component failures prevented its use for the June 25 sample period.

The LIDAR continuously performed vertical scans on the upwind and downwind sides of the field, horizontal scans over the field, and calibration stares throughout tillage observation periods. Lines of approximate vertical scan locations are shown in Fig. 1 by the dashed lines emanating from L1 and L2; horizontal scans moved between upwind and downwind vertical scan positions at 0.75 deg from horizontal. Vertical scans started at 0.75 deg and extended up to between 15 deg and 45 deg. The maximum vertical angle varied between sample periods but was usually $\le 25\mathrm{deg}$. Images resulting from vertical scans were monitored throughout each sample period to ensure that the maximum vertical extent of the plumes was entirely captured; modifications to the maximum vertical extent were made as needed. The LIDAR beam was about 10 m agl at the closest edge of the fields at 0.75 deg in elevation. The beam was kept at or above this level due to eye safety concerns. The LIDAR system did not measure plumes below this level and, therefore, may underestimate PM flux.

A calibration stare refers to short periods (60 to 120 s) when the LIDAR beam is held adjacent to an upwind tower with collocated point sensors. Calibration stares were routinely performed throughout the sample period at 10 to 20 min intervals. In postprocessing, LIDAR return signals collected during calibration stares were calibrated to PSD and $V_k$ measurements. The process used to accomplish this is described in detail by Refs. 23 and 24 and will now be summarized. The calibration process is illustrated in Fig. 2 and was carried out through the following steps:

Fig. 2

Process diagram for LIDAR PM calibration algorithm.

This calibration method allows a scanning LIDAR to estimate ${\mathrm{PM}}_k$ concentrations surrounding an area/source of interest at a much finer spatial scale than possible with point sensors.

2.3.

Emission Calculation Methods

The ${\mathrm{PM}}_k$ data from the point sensors and LIDAR were used to the estimate $\eta$ of the conservation tillage CMP in this study. The point sensor ${\mathrm{PM}}_k$ concentrations were coupled with an air dispersion model through inverse modeling. In typical air dispersion model applications, a source ER is supplied to a model which then calculates the resulting concentration ($C_{\text{sim}}$) at a given receptor location. Inverse modeling involves using a dispersion model and concentrations measurements around a source activity ($C_{\text{downwind}}$, $C_{\text{upwind}}$) in order to estimate the observed ER ($Q_{\text{meas}}$). An initial ER ($Q_{\text{sim}}$) is supplied to the model to calculate $C_{\text{sim}}$, then the following equation is used to calculate $Q_{\text{meas}}$:

If the model used has a proportionally linear response in $C_{\text{sim}}$ to changes in $Q_{\text{sim}}$, the ratio ${(C/Q)}_{\text{sim}}$ is a scalar value independent of the given $Q_{\text{sim}}$ value, i.e., there are no local maxima or minima that might influence the resulting value of $Q_{\text{meas}}$.

AERMOD (American Meteorological Society and U.S. EPA Regulatory Model), a Gaussian air dispersion model that estimates $C_{\text{sim}}$ at a given receptor point based on meteorological conditions, source strength, and the horizontal and vertical distances of the receptor from the source, was selected to perform the inverse modeling estimation of $Q_{\text{meas}}$.²⁷ It is an air dispersion model currently recommended for regulatory modeling by the U.S. EPA. It operates in 1 h time steps, has a proportionally linear response in $C_{\text{sim}}$ to changes in $Q_{\text{sim}}$, and assumes steady-state conditions, continuous emissions during a time step, conservation of mass, and $C_{\text{sim}}$ resulting from multiple sources are additive. Pollutant distribution is modeled as Gaussian in the stable boundary layer in both the horizontal and vertical directions. In the convective boundary layer, horizontal dispersion is modeled as Gaussian while vertical pollutant distribution is modeled as bi-Gaussian. The spatial resolution of $C_{\text{sim}}$ is controlled by the user through discrete and/or gridded receptor points. The commercially available user-interface AERMOD View from Lakes Environmental, Inc. (Waterloo, Ontario, Canada), with AERMOD version 13350, was employed.

Onsite measured wind speed, wind direction, temperature, humidity, and solar radiation were used by AERMET, the meteorological preprocessor for AERMOD, to create both surface and elevated meteorological input files. Wind, temperature, and humidity data were used from the upwind meteorology tower dataset, with wind speed from 9.7 m agl and temperature and relative humidity from 2.5 m agl. Incoming solar radiation was measured at the AQT location. Percent cloud cover was set to zero based on visual observations during the measurement periods.

The land-use classification on all sides of the site was cultivated land. Values provided to AERMET for midday albedo and Bowen ratio of 0.18 and 1.5, respectively, were suggested as average summer values under dry conditions for a fallow agricultural field in Ref. 28. The surface roughness length ($z_0$), also required by AERMET, was calculated based on wind profile measurements at the upwind meteorological tower using the following equation which relates wind speeds ($u_1$, $u_2$ in $\mathrm{m}·{\mathrm{s}}^{-1}$) at two heights ($z_1$, $z_2$ in m) and was derived from the integrated logarithmic wind speed profile equation:

A least sum of squares of residuals methodology was used to determine the value of $z_0$ that best fit measured wind speeds at the higher elevation of two paired wind speed time series over the study period. A $z_0$ value of 0.02 m was calculated as the arithmetic average of the values that best fit six pairings of wind speeds measured at 2.5, 3.9, 6.2, and 9.7 m agl, i.e., 2.5 and 3.9 m, 2.5 and 6.2 m, 2.5 and 9.7 m, 3.9 and 6.2 m, 3.9 and 9.7 m, and 6.2 and 9.7 m. The cup anemometer at 15.3 m malfunctioned during this deployment, rendering the data unusable for this analysis. The upwind location was selected for this analysis as the downwind tower was removed on June 12 to support another study nearby. The AQT was also removed from the southern edge of field 5 at the same time.

Tillage operations were modeled as ground level area sources with initial plume heights of 0 m and areal extents equal to the actual tilled portions of the field. Most operations covered all or most of the fields within a sample period, but some, such as the break down in-field borders operation, were intended to only work a small portion of the field surface. Tilled areas and sampler locations were measured using a hand-held GPS unit. The $Q_{\text{sim}}$ values for each modeled operation were based on a preliminary average ER value across all tillage operations from Ref. 11 of $8.6\mu \mathrm{g}·{\mathrm{s}}^{-1}·{\mathrm{m}}^{-2}$ per operation per pass multiplied by the number of passes over the field within a sample period. Sources were activated or deactivated in hourly time steps throughout a simulated sample period according to the tractor operation times as monitored by onsite personnel. Discrete receptors were set at each sampling location to yield $C_{\text{sim}}$ for inverse modeling comparisons. Uniform Cartesian receptor grids at 2 m agl and 15 m spacing between points were set from upwind sampling locations to several hundred meters downwind of the fields to visualize predicted plume movement, shape, and concentration. Hourly $C_{\text{sim}}$ values were averaged over the modeled sample period. Modeled plume edge effects were avoided by eliminating those locations with $C_{\text{sim}}$ less than 10% of the maximum $C_{\text{sim}}$, adapted from suggestions by Ref. 29, from emissions calculations.

The second ER and EF calculation approach was a mass balance applied to the LIDAR ${\mathrm{PM}}_k$ data. Assuming uniform background aerosol levels, average upwind concentrations were subtracted from concentrations in and around detected plumes in the downwind vertical scans. The difference was multiplied by the component of the wind perpendicular to the beam to calculate the horizontal flux of PM through the downwind vertical scanning plane. Fluxes were summed across the vertical plane, averaged over the length of the sample period, and then divided by the size of the tilled area to calculate the mean EF of ${\mathrm{PM}}_k$ from the field surface. The EF was further divided by the total tractor time to calculate the mean ER of each operation. This method of calculating ERs and EFs using LIDAR is described in detail in Ref. 30.

Vertical profiles of PM mass concentration, horizontal wind speed, and wind direction are required to use the mass balance approach. The PM profile was provided by the LIDAR ${\mathrm{PM}}_k$ data. Profiles of wind speed were calculated using the wind speed power law, as given by Ref. 31,

Wind direction over the vertical profile was assumed to be constant. Though wind direction is known to change in a vertical profile, the influencing factors may be complex and the magnitude and direction of change highly variable. Therefore, in the absence of measured data, the assumption that wind direction did not change with increasing elevation over the 250 m profile was used.

3.1.

Particulate-Matter Concentration Measurements

A total of 296 filter samples were collected: 116 ${\mathrm{PM}}_{2.5}$ (39%), 116 ${\mathrm{PM}}_{10}$ (39%), and 64 TSP (22%). Calculated ${\mathrm{PM}}_{2.5}$ concentrations based on filter catch ranged from 23.2 to $3244.9\mu \mathrm{g}·{\mathrm{m}}^{-3}$; ${\mathrm{PM}}_{10}$ concentrations ranged from 38.1 to $1458.4\mu \mathrm{g}·{\mathrm{m}}^{-3}$; TSP concentrations ranged from 73.6 to $2276.9\mu \mathrm{g}·{\mathrm{m}}^{-3}$. The average method detection limit (MDL, $n=13$), calculated based on sample period duration, the targeted flow of $5.0\mathrm{L}·{\min }^{-1}$, and the minimum detectable difference between pre- and post-test filter weights of $5\mu \mathrm{g}$, was $6.6\pm 4.9\mu \mathrm{g}·{\mathrm{m}}^{-3}$ and the median was $4.3\mu \mathrm{g}·{\mathrm{m}}^{-3}$, with a range of $2.3\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for a run length of 7.3 h to $17.3\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for a run length of 1.0 h.

Of the 296 filter samples collected, 98 (33%) did not pass quality analysis (QA) checks applied to the dataset. QA checks included visual inspection of filter surfaces, sample log inspection for noted problems, OPC time series examination for contamination (used mostly at upwind sites), concentration consistency across sampling locations, and concentration comparisons among ${\mathrm{PM}}_{2.5}$, ${\mathrm{PM}}_{10}$, and TSP at each sample location. In-depth descriptions of the QA checks are found in Ref. 15. An investigation into the cause(s) of this high rate of failure was conducted and a summary of conclusions is provided in the following paragraph. A large number of failures of near-source, downwind samples relative to the total number of downwind samples collected were found in the sample periods from May 18 to May 20. This, when combined with the results from the investigation into the large number of failures, casts doubt on the validity of the remaining near-source downwind samples from those runs. Therefore, all near-source downwind MiniVol samples for these sample periods were removed from MCF and ER calculations, rendering the upwind and far-source downwind samples that passed QA unusable for estimating PM emissions. The filter dataset used to calculate ERs and EFs totaled 131 samples (44%). Concentration ranges for this dataset were 26.7 to $149.8\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for ${\mathrm{PM}}_{2.5}$, 47.4 to $489.4\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for ${\mathrm{PM}}_{10}$, and 102.9 to $1,896.9\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for TSP. The size fraction distribution of filters used to estimate emissions was nearly identical to the total sample set: 51 (39%) were ${\mathrm{PM}}_{2.5}$, 50 (38%) were ${\mathrm{PM}}_{10}$, and 30 (23%) were TSP.

Filters that did not pass QA were found to have been contaminated during one or more of the following stages: sampling, filter handling, and filter storage. Evidence of “particle bounce” was found on many ${\mathrm{PM}}_{2.5}$ and ${\mathrm{PM}}_{10}$ samples collected during May sample periods. Particle bounce occurs when particles that collide with the impactor plate are re-entrained in the airstream and collected on the filter downstream and result in higher reported levels than actually existed. This issue is most likely due to exposing the MiniVol samplers to dust plumes exceeding the maximum recommended exposure level and improper instrument maintenance and cleaning through the May sample periods. Corrective action in the form of inspection after each deployment and cleaning, if needed, was taken during the June sample periods; no issues associated with particle bounce were observed in the second portion of the study. Additionally, some particles were observed on top of and imbedded into the plastic annular ring around the Teflon filter material—the plastic ring is covered by the filter holder assembly during deployment. This was likely due to contamination during onsite filter storage or handling. Efforts were made to minimize this issue throughout, especially during the June sample periods. However, windblown dust did impact the handling and storage area on May 20.

The collected OPC data were used to calculate PSD, $V$, and $V_k$ values. Unlike the downwind MiniVol samplers, the downwind OPCs were not overwhelmed by the dust plumes from the tillage activities—the manufacturer specified range of the OPC of 0 to $3.18\times {10}^8$ particles $m^{-3}$ was never exceeded—and thus provided usable data throughout all sample periods. Background and downwind PSD and $V$ profiles varied throughout the study, as shown in Fig. 4. The particle concentrations in this figure were calculated as the change in number ($N$) per change in natural logarithm of $d_p$ ($dN/d[\ln (d_p)]$) where $d_p$ is the GMD per bin. The particles emitted by the tillage activities were generally large ($d_p>5\mu \mathrm{m}$) and, therefore, strongly dominated the volume and mass contributions of the activity to near-source atmospheric particle loadings. Three to four OPCs were in positions immediately downwind of the field under study in each sample period, with between one and four OPCs in upwind locations.

Fig. 4

Sample period-averaged upwind and downwind particle size distributions (PSDs) as measured by optical particle counters (OPCs) for (a) May 17, strip–till operation, field 5, (b) May 18, chisel operation, field 4, (c) June 5, plant operation, field 4, and (d) June 7, plant and fertilize operation, field 5. Error bars represent the 95% confidence intervals about the average.

OPC time series data at upwind locations were examined for contamination from upwind activities, such as unpaved road traffic. Contamination was found in six of the 12 sample periods, with five of those occurring at the sample site adjacent to the upwind meteorological tower and immediately downwind of an unpaved road. Large, short-duration spikes indicative of contamination were removed from the upwind data in these instances to estimate the background aerosol concentration. In each instance this was performed, the estimated background levels were in very good agreement with those measured by an OPC at a different, uncontaminated upwind location. Filter samples collected at upwind locations with contamination indicated by OPC data were removed from ER estimation.

Those filter samples that passed QA, including the upwind and far-source downwind samples from May 18 to May 20, were used to estimate ${\mathrm{MCF}}_k$ values if collected adjacent to an OPC. Most of the daily average ${\mathrm{MCF}}_{10}$ and ${\mathrm{MCF}}_{\mathrm{TSP}}$ values were within the expected range of 1 to $3\mathrm{g}·{\mathrm{cm}}^{-3}$. However, the daily average ${\mathrm{MCF}}_{2.5}$ values were much larger than expected, with individual values ranging from 3.2 to $28.2\mathrm{g}·{\mathrm{cm}}^{-3}$, having a mean of $14.6\pm 3.7\mathrm{g}·{\mathrm{cm}}^{-3}$ and a median of $10.1\mathrm{g}·{\mathrm{cm}}^{-3}$. For comparison, the densities of pure nickel and mercury are 8.9 and $13.5\mathrm{g}·{\mathrm{cm}}^{-3}$, respectively. In past field campaigns, ${\mathrm{MCF}}_{2.5}$ has generally been higher than ${\mathrm{MCF}}_{10}$ and ${\mathrm{MCF}}_{\mathrm{TSP}}$ values, but these ${\mathrm{MCF}}_{2.5}$ values were much higher than those seen before and account for the majority of values above $5\mathrm{g}·{\mathrm{cm}}^{-3}$ reported in Ref. 21. Due to the nonphysically large numbers, the calculated ${\mathrm{MCF}}_{2.5}$ was not used. Instead, the average soil density of $2.65\mathrm{g}·{\mathrm{cm}}^{-3}$ given in the USDA NRCS National Soil Survey Handbook was used as a constant ${\mathrm{MCF}}_{2.5}$ for all sample periods.³² Using a constant ${\mathrm{MCF}}_{2.5}$ may affect the accuracy of calculated ${\mathrm{PM}}_{2.5}$ concentrations, ERs, and EFs and the value of ${\eta }_{2.5}$. However, as will be shown later, the effect on ${\eta }_{2.5}$ was assumed to be small as the values were very close to those of ${\eta }_{10}$ and ${\eta }_{\mathrm{TSP}}$. Table 4 presents the daily ${\mathrm{MCF}}_k$ values used to convert $V_k$ calculated from LIDAR and OPC measurements to ${\mathrm{PM}}_k$.

Table 4

Mass conversion factors (MCFs) used to convert optical particle measurements to mass concentrations for each sample day and averaged for the whole campaign. Error values represent the 95% confidence interval for n≥3. A constant MCF2.5 value equal to the average density of soil was used due to nonphysically high values calculated for most of the sample days.32

The cause of the high ${\mathrm{MCF}}_{2.5}$ values is unknown. No significant differences in ${\mathrm{PM}}_{2.5}$ chemical composition were observed between sample periods with higher and lower ${\mathrm{MCF}}_{2.5}$ values. While PSDs varied between sample periods, no trends in PSDs sufficient to explain high/low ${\mathrm{MCF}}_{2.5}$ groupings were observed. Higher average ${\mathrm{MCF}}_{2.5}$ values were not restricted to sample periods in which evidence of particle bounce was found (May 18 to May 20); those filters exhibiting evidence of particle bounce were removed prior to MCF calculations. ${\mathrm{MCF}}_{10}$ and ${\mathrm{MCF}}_{\mathrm{TSP}}$ patterns tended to follow ${\mathrm{MCF}}_{2.5}$ patterns, having correlation coefficients ($r$) of 0.64 and 0.84, respectively, but with much smaller changes in amplitude. Good negative correlations ($-0.69\le r\le -0.64$) were found when comparing all ${\mathrm{MCF}}_k$ values with sample duration. This means that ${\mathrm{MCF}}_k$ tended to increase as sample duration decreased.

One potential explanation consistent with these relationships is contamination during filter handling and storage. If filters were equally contaminated, the greatest effect would be found on those samples with the smallest mass catch, i.e., samplers with ${\mathrm{PM}}_{2.5}$ impactor configurations or shorter sample times. Unfortunately, field and lab blanks were not collected to monitor for and quantify such contamination; this oversight has been corrected in subsequent studies.

An alternate contamination test is a comparison with independent and proximate PM measurements, though conclusions from this test are limited by comparability of sample characteristics. The closest independent monitoring site was a suburban monitoring station operated by the SJVAPCD, which reported 24-h average ${\mathrm{PM}}_{10}$ concentrations on 3 days during which tillage monitoring occurred. The SJVAPCD 24-h average ${\mathrm{PM}}_{10}$ and study site mean background ${\mathrm{PM}}_{10}$ pairs were, respectively, 38 and $38\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for May 18, 34 and $47\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for June 5, and 38 and $163\mu \mathrm{g}·{\mathrm{m}}^{-3}$ on June 11. Differences in sample period (24-h versus 1.5-h to 7.5-h and limited to daylight hours only), location setting and local sources (suburban versus rural), and instrumentation (FRM versus MiniVol) existed between the two datasets and contributed to observed differences. The differences in concentration between the two sites on May 18 and June 5 were within the range of expected values and do not support the sample contamination hypothesis. At first glance, the large difference on June 11 may be interpreted to support this hypothesis. However, all filters collected during this sample period easily passed the visual filter inspection for particle bounce and particles on and/embedded into the annular ring. If filter contamination did occur during this field study, it does not appear to have been consistent based on the comparison with proximate ${\mathrm{PM}}_{10}$ measurements nor evident in the applied QA tests. Therefore, the cause of the high ${\mathrm{MCF}}_{2.5}$ values is unknown and no further data exclusions were made.

Sample period-averaged OPC ${\mathrm{PM}}_k$ data were ranged from 4.3 to $60.2\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for ${\mathrm{PM}}_{2.5}$, 41.2 to $641.1\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for ${\mathrm{PM}}_{10}$, and 95.3 to $\mathrm{3,271.9}\mu \mathrm{g}·{\mathrm{m}}^{-3}$ for TSP. OPC ${\mathrm{PM}}_k$ values at the native 20-s averaging period had much higher ranges across all $k$ as the plumes emitted by the roving tillage activities impacted the point samplers in short bursts, the impact of which was reduced when averaged with intervals measuring lower levels. LIDAR-derived ${\mathrm{PM}}_k$ also had a high variability for the same reason. This is evident in Fig. 5, which presents ${\mathrm{PM}}_{10}$ reported by all three systems at 9 m agl at the downwind tower for the June 18 sample period (14:00–16:10). The dashed black line represents the sample period average ${\mathrm{PM}}_{10}$ value based on the MiniVol filter sample, the red line represents the ${\mathrm{PM}}_{10}$ based on 20 s OPC data, and the blue markers represent the LIDAR-derived ${\mathrm{PM}}_{10}$ for each 0.5-s signal averaging period. The higher temporal sampling frequency of the optical systems shows the timing and magnitude of individual plumes impacting the collocated group not resolved by the single MiniVol sample. The LIDAR beam was not continuously adjacent to the point sensors as it was performing upwind calibration stares and downwind vertical scans, leading to the gaps seen in the LIDAR time series. The LIDAR reported elevated ${\mathrm{PM}}_{10}$ levels when collocated during plume impaction events. It also reported elevated levels when the point sensors did not—this was due to part of a plume being within the 6 m long LIDAR bin but not significantly impacting the tower. Negative ${\mathrm{PM}}_k$ values were occasionally reported by the LIDAR, as seen here. These were artifacts of the optimization in the calibration procedure and are not real—negative PM concentrations are not possible.

Fig. 5

Time series of ${\mathrm{PM}}_{10}$ concentrations as reported by the collocated OPC, MiniVol filter sampler, and LIDAR at 9 m agl on the downwind side of the tillage activity for the June 18 sample period (14:00–16:10).

While the scanning LIDAR was at a disadvantage compared to the point sensors in monitoring PM continuously at a given location in Fig. 5, a significant advantage of the LIDAR over the point sensors is its ability to monitor plumes over a line or area. For instance, the data shown in Fig. 6 were collected in a stare adjacent to the downwind tower (location indicated by the solid black line at constant range) over approximately 1 min of the sample period in Fig. 5. The LIDAR detected multiple, highly concentrated plumes at varying distances, only one of which appears to impact the tower. The bottom plot shows the average ${\mathrm{PM}}_{10}$ concentration with range during this time.

Fig. 6

(a) LIDAR-derived ${\mathrm{PM}}_{10}$ in a time versus distance from the LIDAR concentration map and (b) a time series average concentration versus distance from the LIDAR graph. Data were collected in a stare past the downwind tower during the June 18 sample period.

The vertical scanning profiles used to monitor plumes emitted by the tillage activities allowed measurement of vertical and along-beam horizontal plume extents. Monitored plumes reached elevations up to 150 m agl at the downwind LIDAR scanning plane, though most remained below 100 m. Plume widths also varied. Figure 7 provides an example of two plumes captured in a single vertical scan on June 5 when two different tillage operations were performed in different areas of the field. The plume closer to the LIDAR is lower and denser than the plume farther away.

Fig. 7

Image of plumes in a vertical LIDAR scan on June 5.

Comparisons between ${\mathrm{PM}}_k$ concentrations from MiniVol, OPC, and LIDAR data at upwind and downwind locations were performed for each sample period as a check on the use of the ${\mathrm{MCF}}_k$ and the LIDAR calibration procedure. Accurate estimates of ${\mathrm{PM}}_k$ are necessary for accurate estimates of PM emissions. The MiniVol, as it measures ${\mathrm{PM}}_k$ directly at each point, is assumed to be more representative of the actual ${\mathrm{PM}}_k$ than the OPC and LIDAR. Most inter-instrument comparisons revealed upwind values agreed fairly well, but greater differences were found in comparisons at downwind sample sites. Potential reasons for the similarities and differences at both upwind and downwind point sampler locations will be discussed in the following paragraphs, with an example comparison provided as well.

The similarity of ${\mathrm{PM}}_k$ estimated by the OPC and LIDAR at the upwind site was expected as this is the calibration point in the LIDAR PM retrieval algorithm and both are converted to ${\mathrm{PM}}_k$ through the same ${\mathrm{MCF}}_k$. The downwind sites, however, were not used as direct calibration points, and thus may exhibit greater differences between LIDAR and OPC period-averaged values. On a side note, the proximity of all upwind OPC and LIDAR ${\mathrm{PM}}_k$ demonstrates the selected time interval between LIDAR calibration stares was sufficient to adequately characterize changes in upwind PM. It also supports the assumption made in the flux calculations that upwind ${\mathrm{PM}}_k$ is relatively constant.

Differences in calculated ${\mathrm{PM}}_k$ between instruments may result from a variety of factors, including but not limited to the following: sample volume differences ($\mathrm{OPCs}-1\mathrm{L}·{\min }^{-1}$; $\mathrm{MiniVols}-5\mathrm{L}·{\min }^{-1}$; $\mathrm{LIDAR}-6\mathrm{m}$ bin length $x\sim 1\mathrm{m}$ beam diameter sampled at 10 kHz with data averaged over 0.5 s); sampling frequency at the point sensor location (MiniVol and OPC–continuous; LIDAR–upwind: 3 to 5 min per 15 min, and downwind: 1 to 3 min per 15 min); LIDAR sample timing with respect to plume location (i.e., simultaneous presence of both the LIDAR beam and a transient plume impacting the instrumented tower versus the total time the plumes impacted the tower); and the differences between the ${\mathrm{MCF}}_k$ values calculated at the comparison site and the average ${\mathrm{MCF}}_k$ across all measurement sites used to convert $V_k$ to ${\mathrm{PM}}_k$. LIDAR stare time used in the comparison with continuously running, stationary OPCs, and MiniVols was limited by the need to perform vertical scans for flux estimation. Better agreement is expected on the downwind side with an increasing proportion of time spent adjacent to downwind samplers.

The error introduced in this inter-instrument comparison associated with using a spatially averaged ${\mathrm{MCF}}_k$ was generally less than $\pm 30\%$ during this study based on the 95% confidence intervals (CIs) reported in Table 3. Large differences in ${\mathrm{PM}}_{2.5}$ between the filter- and optical-based methods were observed across all sample periods due to the use of a constant ${\mathrm{MCF}}_{2.5}$ value that was not derived from onsite PM measurements. However, OPC and LIDAR ${\mathrm{PM}}_{2.5}$ levels were close.

An example of an inter-instrument comparison is provided in Table 5 for the June 18 sample period, the period shown in Fig. 5. In this instance, average upwind OPC and LIDAR ${\mathrm{PM}}_{10}$ and TSP concentrations were within 15% and 7%, respectively, of the values reported by the MiniVol. Downwind LIDAR ${\mathrm{PM}}_{10}$ and TSP values were 73% and 64% of OPC levels, respectively, while the OPC ${\mathrm{PM}}_{10}$ was 6% higher than the ${\mathrm{PM}}_{10}$ concentration from the filter sample. The differences at the downwind location were likely caused, in large part, by the relatively small amount of time the LIDAR sampled adjacent to the point sensors. Longer stare periods at downwind locations are expected to yield better ${\mathrm{PM}}_k$ accuracy and are planned for future deployments.

Table 5

Comparison of period average PM mass concentrations as reported by collocated MiniVol filter samplers and OPCs, as well as the adjacent LIDAR bin, at measurement heights of 9 m agl at upwind and downwind tower locations for the June 18 sample period. Error values provided for OPC and LIDAR PMk represent the 95% confidence interval (CI).

3.2.

Estimated Emissions and Control Efficiencies

The ${\mathrm{PM}}_k$ data from the LIDAR, OPCs, and MiniVols were all used to calculate PM emissions using mass balance and inverse modeling techniques. Table 6 provides the results of these calculations, as well as the summed emissions from each tillage management practice. Stated uncertainties are the 95% CIs about the average and have been provided for cases where $n\ge 3$.

Table 6

Average particulate matter (PM) emission factors (EFs) and 95% confidence intervals (CI) estimated for the conventional and conservation tillage management practices. Emissions were calculated through applying a mass balance technique to mass-calibrated LIDAR data and inverse modeling with AERMOD and both filter-based particle sampler (MiniVol) and mass-calibrated OPC data. Control efficiency (η) results are also provided for the conservation tillage management practice. NPO=no plumes observed.

The summed conservation tillage sequence emissions ($E_{\mathrm{ST}}$) consist of the following three passes: strip–till, plant, and herbicide application. The LIDAR did not detect plumes in downwind vertical scans during the herbicide application, as indicated by the no plumes observed designation. However, downwind PM samplers reported small increases in concentrations over background levels during this sample period, leading to small EFs relative to the other operations investigated. The herbicide application operation was performed by a small tractor pulling a spray applicator; as no subsurface disturbance occurred, the only active PM sources were the tractor and implement tires and the spray droplets.

The summed emissions for the conventional tillage method ($E_{\mathrm{CT}}$) include the following 13 passes in order: two break down in-field border passes, chisel, two disk passes, lister, two cultivator passes, roll, plant, fertilizer injection, and two more cultivator passes. The EFs for the break down in-field borders operation have been distributed over the entire field area (10.1 ha) instead of just the area worked (1.0 ha) to represent the emissions over the entire field in the $E_{\mathrm{CT}}$ calculation. As the LIDAR system was unavailable to take measurements during the last sample period for the third cultivator pass, the emissions of the last two cultivator passes were assumed to be equal to the observed emissions for the first two cultivator passes. The inverse modeling method found that the third cultivator pass emitted from 0.6 to 3.6 times as much PM as calculated for the first and second cultivator passes, with an average of 2.1.

Some of the LIDAR-derived EFs were much higher than those calculated by inverse modeling within an operation and PM size fraction, particularly for the chisel, disk 1, and lister passes. Others were not statistically different due to overlapping CIs. $E_{\mathrm{CT}}$ and $E_{\mathrm{ST}}$ EFs from the LIDAR dataset were significantly higher, based on the CIs, than those calculated through inverse modeling in all cases but the ${\mathrm{PM}}_{2.5}$ $E_{\mathrm{ST}}$. The LIDAR- and OPC-based ${\mathrm{PM}}_{2.5}$ EFs were smaller than the MiniVol-based ${\mathrm{PM}}_{2.5}$ EFs in all cases, which is likely to be related to the use of the average soil density as the ${\mathrm{MCF}}_{2.5}$ in place of calculated values as discussed earlier.

While the EFs from published studies are generally not in close agreement, a high range of variability is expected from measurements made under different meteorological and soil conditions, as demonstrated in Ref. 5. Some of the ${\mathrm{PM}}_{10}$ EF values calculated from this study were in agreement with those given by Refs. 5 and 10, such as the cultivate, roll, strip–till, conventional tillage planting, and conservation tillage planting passes. Other EFs were much larger than values previously reported, especially the disk 1, disk 2, chisel, and lister passes. The results from this campaign were, in general, not in as good agreement as previous studies have been.

The $\eta$ values were also included in Table 6. They were calculated using the following equation based on a collection efficiency equation found in Ref. 31 and represents the PM emissions reduction of the conservation tillage compared to the conventional tillage:

The strip–till conservation tillage reduced PM emissions in all size fractions by about 90%. The reduction in total tractor operation time per unit area of field was similar to PM reductions at 84% (see Table 1). The $\eta$ results had a small range of $<6\%$ across methodologies and size fractions, despite large differences in summed emissions between methodologies.

Limitations of the PM sensors and AERMOD contribute to limitations and uncertainties in the estimated EFs. For instance, the scanning LIDAR system did not collect data below about 10 m agl at the range of the fields due to eye-safety concerns, which results in portions of plumes not being sampled and included in EF estimates. Vice versa, the inability to locate point sensors above 10 m limits their ability to characterize the plume depth. In this case, the LIDAR and point sensors complement each other in mapping the emitted plumes, as previously demonstrated by Ref. 6. In addition, measurements at a few points may or may not represent the plume characteristics sufficiently to accurately determine the EF, particularly for a roving source such as in agricultural tillage. This was mitigated by deploying as many samplers at different sites within the downwind plume as possible. Another limitation that was identified by Ref. 11 and was also present in this analysis is AERMOD’s poor simulation of elevated plumes from agricultural tillage, particularly plumes completely detached from the ground.