Wildfires and prescribed fires produce emissions that degrade visibility and are harmful to human health. Smoke emissions and exposure monitoring is critical for public and environmental health protection; however, ground-level measurements of smoke from wildfires and prescribed fires has proven difficult, as existing (validated) monitoring technologies are expensive, cumbersome, and generally require line power. Few ground-based measurements are made during fire events, which limits our ability to assess the environmental and human health impacts of wildland fire smoke.
The objective of this work was to develop and validate an Outdoor
Aerosol Sampler (OAS) – a filter-based air sampler that has been
miniaturized, solar powered, and weatherproofed. This sampler was
designed to overcome several of the technical challenges of wildland
fire monitoring by being relatively inexpensive and solar
powered. The sampler design objectives were achieved by leveraging
low-cost electronic components, open-source programming platforms,
and in-house fabrication methods. A direct-reading
Thirteen OAS were deployed to monitor smoke concentrations downwind from a large prescribed fire. Aerosol mass concentrations were interpolated across the monitoring network to depict smoke concentration gradients in the vicinity of the fire. Strong concentration gradients were observed (spatially and temporally) and likely present due to a combination of changing fire location and intensity, topographical features (e.g., mountain ridges), and diurnal weather patterns. Gravimetric filter measurements made by the OAS (when corrected for filter collection efficiency) showed relatively good agreement with measurements from an EPA federal equivalent monitor. However, the real-time optical sensor (Sharp GP2Y1023AU0F, Sharp Electronic Co.) within the OAS suffered from temperature dependence, drift, and imprecision.
Wildfires and prescribed fires are the largest combined source of
primary fine particulate matter (
Fire regimes have changed during the last century due to changes in climate, land-management techniques, agricultural practices, and industrial development (Westerling et al., 2006). Over the past three decades, wildfires have increased in number, size, and severity (Alves et al., 2000; Miller et al., 2009). This upward trend of wildfire activity is predicted to persist in coming years (Flannigan et al., 2000), meaning biomass burning will have an even greater impact on public and ecosystem health in the future (Spracklen et al., 2009; Yue et al., 2013). One method of wildfire mitigation is prescribed burning, a technique that has increased substantially in recent years.
The Interim Air Quality Policy on Wildland and Prescribed Fires was written by the EPA in 1998 to preserve public health and wellbeing by mitigating air quality impacts from prescribed fires (EPA, 1998). Evaluating the effectiveness of smoke mitigation techniques is challenging, however, because emission and exposure monitoring data are sparse. Conventional instruments for monitoring wildfire smoke are expensive, costing USD 10 000 – 30 000 per unit (Strand et al., 2011). These instruments are large and typically require line power; thus, sampling locations are often limited to areas that are equipped with utility service and accessible by motor vehicle. Thus, few measurements are made during most fire events, which often results in an incomplete representation of the fire's impact on local air quality (Hardy et al., 2001).
Satellite observations of air quality can be used to address ground-based monitoring gaps. Moderate resolution imaging spectroradiometer (MODIS) instrumentation, aboard the Terra and Aqua satellites, yield daily aerosol optical depth (AOD) measurements worldwide. AOD is an integrated extinction of light from the total mass of aerosol present in a vertical column of the atmosphere; thus, AOD includes total aerosol mass at all elevations. Satellite-based aerosol measurements, however, still lack precision and fine spatial resolution and do not quantify air quality specifically at ground level (Lassman et al., 2017). As a result, a need still exists for spatially-resolved measurements of surface air quality in the vicinity of fires.
The objective of this work was to develop and validate
a field-deployable, low-cost (under USD 500)
The Outdoor Aerosol Sampler (OAS) was based upon the Ultrasonic
Personal Aerosol Sampler (UPAS) described previously (Volckens
et al., 2017). The original UPAS was designed as a wearable device
to estimate personal exposure to
The following additions and modifications were made to convert
the UPAS from an indoor personal sampler into an outdoor area
monitor: add a direct-reading
Remote communications were accomplished by adding Short Message Service (SMS) technology, which allowed the OAS to be controlled via cell phone (or any device with internet access) and to report data back to a server. The built-in SMS technology and predesignated communication protocol of a Particle Electron (Particle Industries Inc., San Francisco, CA) was utilized for this purpose. The Electron also features a microcontroller that was integrated into the UPAS circuitry, enabling communication among all components.
OAS sampler in
The online
A Pelican 1020 Micro Case was modified to enclose the OAS and to
protect the unit from adverse weather. The mass-flow sensor
within the unit (Omron model DP6F) is dependent on air density,
which can vary as a function of temperature, humidity, and
pressure. These variables were monitored in real-time using an
atmospheric condition sensor (Bosch BME280) and used to correct
mass-flow readings (Volckens et al., 2017). The UPAS creates
a small amount of heat during operation as a by-product of
battery discharge and pump work. Therefore, pump exhaust was
routed through the case (and out a series of small exit holes on
the underside) to help maintain a temperature inside the case
near ambient. At 2
The OAS's two configurations are shown in Fig. 1.
Supplement Table S1 lists the components added to the UPAS in the development of the OAS. The block diagram in Fig. 2 depicts the basic components of the UPAS (colored grey) and the additional components added to create the OAS (colored blue).
Monte Carlo simulations were conducted to inform the selection of
OAS solar and battery power. These simulations were designed to
predict the probability of OAS power failure as a function of
sampling duration (days), time of year (month), available solar
irradiance, solar cell size, and battery capacity. Solar
irradiance, the Monte Carlo sampled input variable, is defined as
the daily average of observed solar irradiance attuned for solar
cell size (0.014
The simulation accounted for the following operational
parameters: useable battery capacity, stationary solar conversion
efficiency, temperature effects on battery capacity, charging
circuit efficiency, and average OAS power consumption at
2
The solar cell arrangement was designed to be collapsible to maintain a slender profile for easy transportation and shipping. A magnetically coupled bracket that is adjustable for optimum zenith-angle holds the solar cells rigidly in place while in deployment and transportation configuration (Fig. 1). A voltage regulator was added to the OAS battery charge controller to condition electricity from the solar cells to 5 V DC.
Block diagram of UPAS technology with component integration to form the OAS: GSM (Global System for Mobile Communications), PM (particulate matter), GPS (Global Positioning System).
Thirteen OAS units were arrayed in the vicinity of a 6000 acre
prescribed fire (known as the Pargin Mountain fire), with
assistance from the Colorado Department of Public Health and
Environment (CDPHE) and the US Forest Service (USFS). The fire
took place 14
A map of sampler locations and the area burned is shown in
Fig. 3. The USFS monitored air quality during the prescribed burn
by placing instruments at location 9 (E-SAMPLER, Met One
Instruments, Grants Pass, OR) and location 1 (E-BAM, Met One
Instruments, Grants Pass, OR). Two OAS were co-located with each
USFS monitor at these locations. For the duration of the fire,
each OAS was programmed to sample PM
Location of monitoring equipment with respect to the prescribed fire. OAS (yellow circles), US Forest Service equipment (blue triangles), prescribed fire (shaded black area with red outline). Map elements created using ggplot (Kahle and Wickham, 2013).
Solar energy conversion efficiency was evaluated for each OAS and across all sampling periods. Data from the voltage/current sensor on the OAS circuit board were used to determine the ratio of solar energy delivered to OAS batteries relative to available solar irradiance. Hourly irradiance measurements were provided by a weather station (PRAWS 5) located on Pargin Mountain during the month of September 2016.
Following the Pargin fire deployment, we verified the accuracy and
precision of the OAS with respect to time-integrated
Filters were contained in individual filter-keepers, inside sealed
plastic bags, for both transportation and storage. Filters were
placed in an equilibrium chamber for at least 12 h before pre-
and post-weighing and discharged on a polonium-210 strip for
a minimum of 15 s prior to weighing on an analytic microbalance
(Mettler Toledo XS3DU;
Descriptive statistics were calculated for all mass concentration data, including identification of outliers, which were primarily cases when the OAS was explicitly known to have malfunctioned (stopped sampling, underflowed, etc). For OAS performance comparison with respect to USFS equipment, measurements were considered valid if the sampler spanned more than 75 % runtime and flow remained within 20 % of desired control. Limit of detection for gravimetric measurements was defined as the average blank mass gain plus three times the SD of the change in blank mass. Limit of quantification was defined similarly but using five times the SD in blank mass change.
Probability of OAS power failure evaluated for various power designs
(number of solar and battery cells) as a function of the number of continuous
sampling days. Colors represent the number of solar panels
(0.014
Data analyses were conducted using Excel 2016 (Microsoft Corp.,
Redmond, WA, USA), Matlab 2015 (The MathWorks Inc., Natick, MA)
and R 3.3.2 (R Core Team, Vienna, Austria). Spatial interpolation of
prescribed fire sampling results was based on ordinary kriging
methods and plotted using gstat in R (Benedikt et al., 2016;
Pebesma, 2004). Model interpolations were constrained to an area
(search radius) of no more than 3
Several key modifications helped streamline the use of the OAS in the field. The replaceable filter cartridge (Fig. S5b in the Supplement) eliminated the need for direct filter handling in the field (during change outs), which reduced the risk of contamination and also aided in sample transport. A threaded aluminum inlet cap (Fig. S5) sealed the filter cartridge in place and provided a rough inlet to protect against intrusion by small insects. The added costs (bill of materials) to convert the UPAS into the OAS totaled USD 183 for a single unit (Table S1).
View from Chimney Rock, facing west on the morning of 18 September 2016 when smoke is visible in several valleys (Photo courtesy of Columbine Wildfire Management). OAS locations depicted by yellow markers. Visible smoke is observed around several OAS while other locations appear to be smoke free. Red arrows indicate location of prescribed fire operations.
Power failure probabilities (representing the chance the OAS will
experience power failure before the conclusion of a given number of
consecutive sampling days) are shown Fig. 4 for the month of
April. These results demonstrate the trade-off between run duration
and the quantity of solar panels and battery cells inside the OAS
(Fig. 4). Based on these simulations, a final design consisting of
three solar panels (0.042
Power failure probabilities for the final OAS design are shown in Fig. S1 for six months of the year. The simulation results suggest that the OAS can achieve three full days of continuous sampling during late fall and winter, greater than four days in spring months, and a full week of continuous sampling in summer. The internal battery, when fully charged, allows for two full days of continuous sampling for all months of the year, regardless of the availability of solar power.
The Monte Carlo simulation was based on data from Fort Collins, Colorado between 2011 and 2015. Weather patterns, a large driver of available solar irradiance, are expected to vary by region. Thus, these simulation results are not generalizable beyond the Colorado Front Range. Further, the simulation selected random days within the specified month (i.e., blocks of consecutive days were not sampled) for any of the 4 years. Random selection of days can attenuate the effect of large weather systems, which may also impact OAS runtime.
An early morning photo (Fig. 5) taken 18 September 2016 from Chimney Rock National Monument facing west shows the location of samplers in the OAS network during the Pargin burn. Smoke from the smoldering fire (red arrows) is observed down slope in the valley bottoms. An image captured on the morning of 17 September from a relay station (Fig. 6), 2.4 km northwest of location 9 facing east, depicts the OAS network from a second view point.
The well characterized Pallflex Fiberfilm line of filters (originally intended for use within the OAS) was discontinued during this project; an alternative filter (Tisch PTFE) was selected. Tisch PTFE filters were selected because they exhibit a relatively low pressure drop and are comprised of hydrophobic polytetrafluoroethylene, which is less susceptible to organic vapor adsorption artifacts than other fibrous filter materials (Kirchstetter et al., 2001; Mader and Pankow, 2001). Prescribed fire sampling results suggested a non-ideal collection efficiency for accumulation-mode aerosol using the Tisch PTFE filters. Subsequent laboratory tests, using a previously described protocol (Cardello et al., 2002), confirmed the relatively low collection efficiency of these filters (results shown in Fig. S2). The estimated mass collection efficiency of these filters was 66.7 % (see Supplement for a description of the method to evaluate filter collection efficiency), assuming a size distribution for an unaged biomass burning aerosol (Sakamoto et al., 2016). Mass concentration data reported here have been corrected for filter collection efficiency.
Smoke covering valley floors caused by an inversion on the morning
of 17 September 2016. Photo taken from relay station 2.4
A total of 61 OAS deployments were made over the nine-day
prescribed fire. Seventeen of the 61 deployments failed to
complete an intended measurement. Approximately half of these
failures (Fig. S6,
Data mapping and interpolation techniques (ordinary kriging) were
used to investigate the spatial and temporal evolution of
ground-level
Factors that may affect sensor performance include, but are not limited, to changes in aerosol size and refractive index, ambient humidity, and ambient temperature. Biomass burning aerosols are known to span a range of particle sizes and refractive indices; these properties can also change over time due to aerosol processing in the atmosphere (Vakkari et al., 2014). Increases in humidity may lead to overestimation of (dry) aerosol mass concentration due to water uptake by hygroscopic particles. An ambient relative humidity of 60 % is considered a lower threshold for water uptake to begin affecting nephelometer response (Chakrabarti et al., 2004); this level was exceeded for 38 % of the sampling time during the Pargin fire. However, relative humidity rarely exceeded 70 % during this period (7 % of the time). Published growth factors for biomass burning aerosol are relatively low at 70 % humidity (Rissler et al., 2006), indicating that water uptake from particle hygroscopicity (and, thus, sensor response) was probably not substantial during the Pargin fire. The effect of temperature on sensor response can be manifested by influencing particle size via gas-particle partitioning and by affecting the sensitivity/response of the photodiode and photodetectors inside the sensor itself. The latter effect is shown in Fig. S4.
A time series depicting variability of daily
The real time optical sensor (Sharp GP2Y1023AU0F) integrated with
the OAS was determined to be unreliable for measuring PM in an
outdoor setting. The sensor was affected by meteorological
variables and inconsistent drift patterns, which precluded the use
of this sensor as a trigger for the gravimetric sampler. The Sharp
sensor's output voltage with respect to ambient temperature is
displayed in Fig. S4, demonstrating a strong linear
trend. A second issue with the Sharp sensor was baseline drift,
which spanned as much as 50
Maps illustrating spatial concentration gradients and the temporal
evolution of fire emissions for
A performance comparison between the OAS and E-BAM (co-located at
Arboles Fire Station, location 1) is shown in Fig. 9. The E-BAM
measures PM mass concentrations using Beta attenuation and has
been shown to agree closely with FRM monitors (Trent,
2006). A Deming regression of the E-BAM and OAS yields a slope of
1.01 and an intercept of
Solar power harvested by the OAS was compared to solar irradiance data for the duration of the prescribed burn sampling. On average, 6.7 % of incident solar energy was converted into useful battery power by the OAS. Solar conversion efficiency measured in the field was slightly less than the 7.5 % efficiency input to the Monte Carlo simulation. On average, however, the solar circuit added an average of 11 h runtime to the OAS during a given 24 h period.
For the post-fire validation experiments, the OAS and PEM samplers
reported good agreement for sodium chloride aerosol measured in
chamber tests (
Prescribed fire summary of air quality at all locations for all dates sampled by both OAS and Forest Service equipment (E-BAM and E-SAMPLER). OAS concentrations displayed fit the following criteria: sample more than 75 % of 24 h sampling period and remain within 20 % of desired flow control at all times.
Performance of the Outdoor Aerosol Sampler relative to an E-BAM
federal equivalent Monitor (meets US-EPA requirements for Class III
designation for
The 13 OAS samplers deployed in a network on the prescribed fire captured high concentration gradients resulting from smoke. Actual gradients, however, may have been even stronger than what was measured. One method of improving the spatial resolution of the network would be to deploy more OAS units. However, given the terrain features for the Pargin burn, only about 15 OAS units could be feasibly deployed by a single person in a 24 h period.
The assumption of a fixed OAS power consumption did not allow the
simulation to account for high filter loadings and the associated
increased OAS power consumption. High aerosol concentrations
(i.e.,
A hindrance of the remote communications method used is the limited availability of the Particle hosted service. The service is only available while the Particle Electron is online, resulting in increased power consumption if communication is to be maintained at all times. Another issue is the execution frequency permitted by Google Scripts. Google Scripts is a free service; however, execution frequency limits data collection to once per hour. A possible solution addressing the limited availability of the Particle web page would be the use of an interrupt queue. This prompt would significantly reduce server time and power consumption. A personal server designed for OAS communication would alleviate issues associated with data collection frequency and simplify data archiving.
The Sharp sensor suffered from unpredictable drift issues,
rendering the real-time measurements unreliable. Although
post-sampling calibration (i.e., normalizing the sensor data to
the 24 h filter mass concentration) would alleviate some of this
error, the baseline drift issue (Fig. S7) would still produce
a substantial bias in reported PM concentrations. Possible OAS
improvements include replacing the real-time sensor (Sharp) with
a more reliable PM sensor. Low cost PM
Reference instruments used to assess outdoor air quality tend to be
expensive and bulky. This project developed and tested an Outdoor
Aerosol Sampler (OAS) that is compact, weatherproof, battery
powered, and designed to approach reference-quality measurements of
Thirteen OAS were deployed around a large prescribed fire in
southern Colorado to evaluate its effectiveness as a smoke
monitoring tool. The OAS network provided spatially resolved
measurements in regions where sampling with current
state-of-the-art equipment was not possible. Strong concentration
gradients were observed and likely present due to topographical
features (e.g., mountain ridges) and diurnal weather patterns. At
extremely high concentrations (i.e., 24 h
The cost, independent power capability, and compactness of the OAS provide a practical means for more effective monitoring of smoke from a prescribed burn or wildfire event. The successful demonstration of a low-cost sensor network represents a first step towards providing burn managers, state and federal agencies, and concerned citizens with a better understanding of fire smoke emissions and resulting exposures. The OAS is not only limited to fire events and may also be used for many other applications of outdoor air quality monitoring. At nearly 1/20th the cost of current state-of-the-art field monitoring equipment, the OAS may be deployed in higher quantities under the assumption of fixed fiscal resources. Air quality data at more locations has the potential to enhance the accuracy of exposure models, yielding a more comprehensive estimate of potential human and environmental health hazards from smoke.
A series of CSV files of 24 h concentrations and other sample data (start/stop times, locations, sampled air volumes, run times) from the Pargin fire and post-fire evaluation experiments is available as part of an online Supplement.
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
The authors wish to thank the following individuals for their contributions to this work: Josh Smith for software and firmware development; Nick Good, John Mehaffy, and Christian L'Orange for assistance with field sampling and data analysis methods; John (Jay) Godson (USFS), Sarah Gallup (CDPHE), Pat McGraw (CDPHE), and Ken Helcoski (CDPHE) for their coordination with field sampling. This research was funded in part by a cooperative agreement from the Joint Fire Science Program (16-2-01-3). Edited by: Francis Pope Reviewed by: two anonymous referees