You are on page 1of 23

Masaya volcano, Nicaragua: How effective are low-cost air

quality monitors for identifying SO2 and PM volcanic


signatures in downwind locations?

Whitty, R. C. W.*,1 , Pfeffer, M. A.2 , Ilyinskaya, E.1 , Roberts, t.3 , Schmidt, A.4,5 ,
Barsotti, S.2 , Strauch, W.6 , Crilley, L. R.7 , Pope, F. D.8 , Bellanger, H.9 , Mendoza, E.6 ,
Mather, T.10 , Liu, E.11 , Peters, N.12 , and Taylor, I.13

* Corresponding author: eercww@leeds.ac.uk


1 Institute of Geophysics and Tectonics, School of Earth and Environment, University of Leeds, Leeds, United Kingdom
2 Icelandic Meteorological Office, Reykjavik, Iceland
3 CNRS UMR7328, Laboratoire de Physique et de Chimie de l’Environnement et de l’Espace, Université d’Orléans, Orléans,
France
4 Department of Geography, University of Cambridge, Cambridge, United Kingdom
5 Department of Chemistry, University of Cambridge, Cambridge, United Kingdom
6 Instituto Nicaragüense de Estudios Territoriales (INETER), Managua, Nicaragua
7 Department of Chemistry, University of York, York, UK
8 School of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, UK
9 Department of Geography, University of Montreal, Quebec, Canada
10 Department of Earth Sciences, University of Oxford, Oxford, UK
11 Department of Earth Sciences, University College London, London, UK
12 Department of Electronic and Electrical Engineering, University College London, UK
13 COMET, Atmospheric, Oceanic and Planetary Physics, University of Oxford, Oxford, OX1 3PU, UK

Abstract

Abstract limit of 150 words; currently at 144 words


Masaya volcano, Nicaragua, is one of the largest persistent volcanic polluters worldwide and causes signif-
icant regional damage. This study analyses data from a network of low-cost monitoring pods, which were
a first-attempt at establishing an air quality network. Data were analysed to determine the network’s effec-
tiveness at identifying the presence of volcanic plume. Modelled meteorological data and visual analysis
of satellite imagery were used to identify plume presence at measurement stations. The two measurement
stations within 4 km of the volcanic source point were reasonably effective at recognising plume presence,
reporting a gas-to-particle volcanic signature 61% of the time that the plume was at El Panama and 41% of
the time at Rigoberto. The instruments in the far-field (16 km from source) were less effective at recogniz-
ing plume presence from gas-to-particle signatures, though concentrations of particulates increased on days
when the plume was present.

1 1 Introduction 11 with intense degassing activity [McBirney, 1956;


12 Stoiber et al., 1986]. The latest and current de-
13 gassing crisis started in 1993 and has resulted
2 Masaya is an active basalt volcanic complex in
14 in large volumes of volcanic gas emissions into
3 Nicaragua, Central America. It has a near con-
15 the atmosphere [Rymer et al., 1998; Burton et al.,
4 tinuous history of pit crater formation, sporadic
16 2000; Williams-Jones et al., 2003; Mather et al.,
5 lava lake activity and degassing as far back as
17 2006b] with SO2 flux ranging from 120 to 2680
6 the 1700s [Rymer et al., 1998]. Santiago pit
18 ton/day (Figure 1).
7 crater, initially formed around 1858-1859 [McBir-
8 ney, 1956], is the location of current volcanic ac- 19 Masaya volcano has a subdued topography,
9 tivity and has undergone five periods of lava lake 20 situated at 635 metres asl, and this often re-
10 development and multiple phases of gas crisis 21 sults in grounding of the volcanic plume in the
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

Figure 1: SO2 emissions from Santiago crater, Masaya volcano, from 1979 to 2020. Black dashed line indi-
cates initiation of gas crisis in May 1993. Grey shaded area indicates the AQMesh downwind measurement
period (this study) from February to August 2017. SO2 emissions data sourced from Stoiber et al. [1986];
Burton et al. [2000]; Delmelle et al. [2002]; Williams-Jones et al. [2003]; Galle et al. [2003]; Mather et al.
[2006a]; Nadeau and Williams-Jones [2009]; Aiuppa et al. [2018]; Global Volcanism Program [2021].

22 low atmospheric boundary layer [Delmelle et al., 53 background ambient atmosphere, as well as its
23 2002]. Prevailing easterly winds mean the plume 54 well-recognised environmental and air quality
24 is commonly moved towards the Las Sierras, 55 impacts [Lambert et al., 1988; Cadle et al., 1971;
25 an area higher (925 metres asl) than the crater 56 Loughlin et al., 2012; Schmidt et al., 2015]. Ex-
26 summit, causing damage to vegetation (includ- 57 posure to SO2 can result in irritation and inflam-
27 ing cultivated crops), machinery and buildings 58 mation of the eyes and the upper respiratory tract
28 over a large area downwind of the crater [Bax- 59 [Longo et al., 2008], and long-term health conse-
29 ter et al., 1982; Delmelle et al., 1999; van Manen, 60 quences can result from relatively short-term ex-
30 2014; Williams-Jones and Rymer, 2015]. Volcanic 61 posure times [Pohl, 1998; Miller, 2004]. Popu-
31 plumes which remain in the lower atmosphere, 62 lation sub-groups including asthmatics, children,
32 such as emitted from Masaya volcano, are com- 63 and respiratory- or cardiac-compromised individ-
33 monly comprised of a highly complex and evolv- 64 uals are particularly vulnerable to exposure to
34 ing mixture of both volcanic and atmospheric 65 SO2 [CRI, 2004; ATSDR, 1998]. Asthmatic indi-
35 gases as well as primary and secondary aerosol 66 viduals who are exposed to mass concentrations
36 particles, dust and ash [Oppenheimer and McGo- 67 of 1310 µg/m3 SO2 for three minutes are likely to
37 nigle, 2004; Pfeffer et al., 2006b; von Glasow et al., 68 have an induced respiratory attack [Balmes et al.,
38 2009; Langmann, 2014; Mason et al., 2021]. Vol- 69 1987; ATSDR, 1998].
39 canic plumes released into the lower atmosphere 70 Once in the atmosphere, SO2 is affected by
40 can have a large impact on air quality, the envi- 71 chemical and physical processes, including reac-
41 ronment and human and animal health across lo- 72 tions with liquid and solid suspended particles
42 cal to regional areas [Tam et al., 2016; Schmidt 73 and gas-phase reactions, leading to conversion
43 et al., 2015; Mather, 2015; Ilyinskaya et al., 2017; 74 of SO2 to aerosols [Stockwell and Calvert, 1983;
44 Whitty et al., 2020; Carlsen et al., 2021a,b; Ilyin- 75 Delmelle et al., 2002; Allen et al., 2002]. The life-
45 skaya et al., 2021]. The injection of the plume into 76 time of SO2 in the troposphere is usually consid-
46 the low atmosphere means that the plume can be- 77 ered to be in the range of one - three days to a
47 come trapped and "ground", causing exposure of 78 week [Rotstayn and Lohmann, 2002; Allen et al.,
48 the land-surface to high concentrations of toxic 79 2002; Pfeffer et al., 2006a; Pattantyus et al., 2018],
49 gas and aerosols. 80 with the rate of SO2 conversion depending on the
50 Sulfur dioxide is often the focal point of gas 81 relative humidity, temperature, interactions with
51 emission monitoring at volcanoes due to its high 82 clouds and the availability of oxidants [Oppen-
52 concentration in volcanic plumes relative to the 83 heimer et al., 1998; Saxena and Seigneur, 1987].

Page 2
84 Through a number of reaction pathways (includ- 142 by Delmelle et al. [2002] found background con-
85 ing oxidation with the hydroxyl radical, OH, and 143 centrations of SO2 in the range of < 2 ppb (parts
86 with hydrogen peroxide, H2 O2 , and O3 ), SO2 is 144 per billion) to the east of Masaya volcano in up-
87 gradually converted to sulfate aerosol, H2 SO4 , 145 wind locations in 1998 and 1999. The high-
88 [Stockwell and Calvert, 1983; Allen et al., 2002], 146 est SO2 levels were measured to the west in the
89 which is a dominant component of volcanic PM2.5 147 area within 4 km of Masaya volcano, with con-
90 [Pattantyus et al., 2018; Tam et al., 2016]. 148 centrations of up to 230 ppb and descriptions
91 Volcanic aerosols can be monitored in real-time 149 of the local vegetation as "devastated" [Delmelle
92 by measuring the particulate matter (PM) in vol- 150 et al., 2002]. SO2 concentrations in the Las Sier-
93 canic plumes. Particulate matter is commonly 151 ras highlands were in the range of 100 ppb in
94 sub-divided into size categories of PM1 , PM2.5 and 152 1999 [Delmelle et al., 2002]. As well as impact-
95 PM10 (PM with diameters <1 µm, <2.5 µm and 153 ing the local vegetation, Masaya’s volcanic plume
96 <10 µm, respectively). This categorisation into 154 has a strong interaction with metal structures,
97 cumulative size modes is important because parti- 155 particularly the roofs of buildings, which consis-
98 cles of different sizes can have varying health im- 156 tently have to be replaced or painted every six
99 pacts, with smaller particles having a larger sur- 157 months due to rapid corrosion [Baxter et al., 1982;
100 face area for the absorption of toxic chemicals as 158 Delmelle et al., 2002; van Manen, 2014].
101 well as a greater efficiency at physical transloca- 159 Masaya volcano is monitored by Instituto
102 tion from the respiratory tract to other areas of 160 Nicaragüense de Estudios Territoriales (INETER).
103 the body [Schlesinger et al., 2006]. For exam- 161 The national park which surrounds the volcanic
104 ple, fine air pollution in the PM2.5 size category 162 complex is occasionally closed to public visitors
105 has been found globally to cause 3% of mortal- 163 and tourists during periods of extreme degassing
106 ity from cardiopulmonary disease and 5% mor- 164 [Duffell et al., 2003]. As far as the authors are
107 tality from cancer of the bronchus, lung and tra- 165 aware there is no established procedure in place
108 chea [Cohen et al., 2005]. The chemical composi- 166 for the local authorities to communicate the haz-
109 tion of volcanic PM is highly heterogeneous, with 167 ards of the volcanic plume to the several thou-
110 common chemical species including sulfates (pri- 168 sands of people who live in the downwind areas.
111 mary emissions or formed via oxidation of sulfur 169 The 2016 - 2019 "Unseen but not unfelt: re-
112 gases) [Cadle et al., 1971; Stockwell and Calvert, 170 silience to persistent volcanic emissions" (UN-
113 1983; Allen et al., 2002; Mather et al., 2003; Lang- 171 RESP) research project was established to inves-
114 mann, 2014] and halides, with an array of met- 172 tigate how resilience is built to living with per-
115 als and metalloids including environmentally- 173 sistent volcanic emissions and the environmental
116 harmful species such as lead and cadmium [Lang- 174 pollution hazard they pose, with Masaya as the
117 mann, 2014; Longo, 2013; Ilyinskaya et al., 2017; 175 case study. Here we examine data collected by
118 Mason et al., 2021; Ilyinskaya et al., 2021]. Vol- 176 the UNRESP project using a relatively low-cost in-
119 canic PM may also include particles derived from 177 strument network installed in downwind commu-
120 industrial and transport sources, ambient matter, 178 nities. We analyse the data from the instrument
121 and fine wind-blown mineral dust [Lim et al., 179 network to determine periods when there was a
122 2012; Holgate, 2017; Tam et al., 2016; Butwin 180 correlation between SO2 and particulates, indicat-
123 et al., 2019]. It is estimated that exposure to am- 181 ing the presence of volcanic plume, and under-
124 bient PM2.5 globally accounts for more than three 182 take a study of the concentrations of different par-
125 million premature deaths each year [Lim et al., 183 ticulate size fractions. The effectiveness of the in-
126 2012], and is especially linked to increases in mor- 184 strument network for monitoring the air quality is
127 tality resulting from cardiovascular and respira- 185 analysed, as well as an investigation of other po-
128 tory diseases, particularly in vulnerable individu- 186 tential monitoring network options.
129 als [Holgate, 2017]. Exposure to H2 SO4 , a dom-
130 inant component of volcanic PM [Mather et al.,
131 2006a; Ilyinskaya et al., 2017; Mason et al., 2021],
187 2 Methodology
132 can result in irritation of the eyes and respira-
133 tory tract, and damage to lung tissue over longer 188 2.1 AQMesh Pods
134 durations [Schlesinger, 1985; Williams-Jones and 189 AQMesh pods are air quality monitoring systems
135 Rymer, 2015]. 190 which are relatively low-cost, in the range of
136 The persistent SO2 and PM-rich volcanic plume 191 £7000 per pod (quote from ACOEM Air Monitors,
137 emitted from Masaya volcano has led to long- 192 2021). Their configuration can be specified by the
138 term contamination and fumigation of an area 193 purchaser allowing a range of possible monitor-
139 >1200 km2 downwind of the volcano following 194 ing options including a variety of gas species, par-
140 the prevailing wind direction [Delmelle et al., 195 ticulate matter, humidity and ambient noise and
141 2002; Williams-Jones and Rymer, 2015]. A study 196 wind conditions. Gases are measured by elec-

Page 3
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

Table 1: AQMesh pod sensor specifications for Nicaragua installation. Instrument specifications as stated by AQMesh
[2021b].
Sensor Type Units Range Precision Accuracy Lower
Limit
SO2 Electrochemical ppb 0 - 100,000 ppb >0.7 20 ppb <5 ppb
NO Electrochemical ppb 0 - 20,000 ppb >0.9 1 ppb <1 ppb
NO2 Electrochemical ppb 0 - 20,000 ppb >0.85 4 ppb <1 ppb
CO Electrochemical ppb 0 - 1,000,000 ppb >0.8 20 ppb <50 ppb
O3 Electrochemical ppb 0 - 20,000 ppb >0.9 5 ppb <1 ppb
PM Optical Particle µg/m3 0 - 250,000 µg/m3 >0.85 5 µg/m3 0 µg/m3
Counter
Humidity Solid sate % 0 to 100 % >0.9 5% RH 1% RH

197 trochemical sensors (B4 series manufactured by 240 ACOEM Air Monitors, 2021). The specifications
198 Alphasense Ltd). Electrochemical sensors oper- 241 of the sensors used in the AQMesh pods installed
199 ate by diffusing the target gas through a porous 242 in Nicaragua by the UNRESP project are outlined
200 membrane, following which changes in the chem- 243 in Table 1.
201 ical potential are measured by a sensing electrode
244 During an UNRESP field campaign, five
202 [Austin et al., 2006; Mead et al., 2013]. The elec-
245 AQMesh pods were installed in settlements to the
203 trochemical gas sensors and the optional humid-
246 west of Masaya volcano, downwind of the volcano
204 ity sensor are mounted into a base plate which
247 during prevalent wind conditions. One AQMesh
205 allows them to come into contact with the am-
248 pod was purchased by the UNRESP project and
206 bient air. Particulate matter is measured by an
249 four were rented for the duration of the mea-
207 OPC-N2 optical particle counter (manufactured
250 surement campaign (rented AQMesh pods sup-
208 by Alphasense Ltd) which uses a laser beam to
251 plied by ACOEM Air Monitors, United Kingdom).
209 detect particles from 0.38 to 17 µm in diameter
252 AQMesh pods were installed at least 1 metre
210 [Crilley et al., 2018]. Particles are assigned into
253 above the ground at sites which were deemed to
211 size fractions of PM1 , PM2.5 and PM10 using an
254 have low likelihood of anthropogenic pollution,
212 embedded algorithm developed by AQMesh, and
255 with the exception of Rigoberto, which was rela-
213 the raw size-resolved data is not available to the
256 tively close to a domestic cooking fire site. The lo-
214 user except upon request [AQMesh, 2021a]. The
257 cations of the AQMesh measurement sites and the
215 AQMesh pods are fitted with a small pump to
258 prevailing plume trajectory are indicated in Fig-
216 pass the ambient air through the OPC-N2 which
259 ure 2.
217 is fitted internally at the top of the instrument.
218 The system is housed in an ABS IP65 box with 260 The AQMesh pods were maintained by the
219 a mounting bracket for installation. The dimen- 261 UNRESP project. Technical specifications of
220 sions are 170 mm by 220 mm by 250 mm with 262 the AQMesh pods state that the electrochem-
221 an additional 180 mm height if an antenna is fit- 263 ical sensors require replacement after 2 years
222 ted [AQMesh, 2021b]. The pods weigh between 264 [AQMesh, 2021b], however the electrochemical
223 2 to 2.7 kg depending on sensor and battery con- 265 sensors needed replacement several times during
224 figuration. Power is supplied either by mains 266 the operational period. At El Crucero station, is-
225 power at 9 - 24V or by an internal lithium metal 267 sues with the SO2 sensor and corrosion of the bat-
226 battery pack which requires recharging every four 268 tery connectors required replacement parts to be
227 weeks [AQMesh, 2021b]. AQMesh states that 269 installed on four occasions within the six months
228 the electrochemical gas sensors and optical parti- 270 of AQMesh pod deployment, resulting in the pod
229 cle counter are fully calibrated during the man- 271 being offline for 36% of the measurement time
230 ufacturing process and have an expected lifes- 272 (Table 2). In some instances the internal batter-
231 pan of two years before replacement is neces- 273 ies were not recharged promptly after four weeks,
232 sary [AQMesh, 2021b]. AQMesh pods are pri- 274 resulting in periods of missing data due to lack of
233 marily manufactured for the monitoring of ur- 275 power. The AQMesh pods were designed for mon-
234 ban and commercial environments. The measure- 276 itoring air quality in urban and commercial envi-
235 ments are uploaded automatically via mobile net- 277 ronments. Exposure to a volcanic environment,
236 work to the AQMesh server which allows data to 278 even in reasonably dilute downwind conditions,
237 be downloaded and also to be viewed in tabular 279 likely led to mechanical issues with the pod op-
238 and graphical formats. There is an annual fee of 280 erating systems. A high level of fast-acting cor-
239 £480 for use of the online web server (quote from 281 rosion impacted the pods at all of the measure-
282 ment sites, but especially those in closer proxim-

Page 4
Figure 2: Topographic map of Masaya volcano and the AQMesh sampling sites with heights in metres
above sea level (asl) for sampling sites and Masaya’s Santiago Crater. Upper left inset indicates the wind
rose referring to 948 metres asl for the period February to August 2017, data derived from ECMWF mod-
elled meteorological data. The red shaded area indicates prevailing plume dispersion, graphically pre-
sented from the most frequent (76%) wind direction derived from ECMWF data at 948 metres asl. Lower
left inset indicates geographical position of Masaya volcano. Base topographic map from Krogh [2021].

283 ity to the volcanic source point. Corrosion oc- 306 EPA, 2016]. FEM-designated instruments (Fo-
284 curred both externally to the mounting brackets 307 rum for Environmental Measurements), such
285 (Figure 3A, B) and internally (Figure 3C, D) to the 308 as this SO2 analyser, promote consistency in
286 metal parts of the sensors, computer boards and 309 measurements and ensuring that the instru-
287 battery charging connectors, resulting in long pe- 310 ments are of reference-grade quality [EPA,
288 riods of missing data with sensors offline until re- 311 2016]. The SO2 analyser was installed in an
289 placement parts could be installed. 312 air-conditioned building at El Crucero [Figure 2]
313 with an inlet tube feeding air in from outside.
290 2.2 Instrument Co-Location and Field Calibra- 314 AQMesh pods 712150 and 735150 were installed
291 tion 315 within a few meters of the SO2 analyser inlet.
316 The SO2 analyser was calibrated on return to
292 To test the precision of the AQMesh sensors, 317 the UK and found to have a baseline drift of
293 all pods were co-located in an urban location 318 2 ppb and an underestimation of 18% [Read,
294 [Figure 2] with minimal anthropogenic pollution 319 2018]. The SO2 analyser was verified using a
295 sources, away from volcanic input (eleven days in 320 National Physical Laboratory certified Cylinder
296 July 2017). 321 (Cylinder number: 176433, BOC Ltd) with the
297 To test the accuracy of the SO2 measure- 322 blender set-up within the AMOF COZI Labora-
298 ments, two AQMesh pods were co-located with 323 tory, National Centre for Atmospheric Science
299 a pulsed fluorescence spectroscopy analyser 324 (NCAS, https://amof.ac.uk/laboratory/carbon-
300 (Thermo Scientific 43i) for two days in December 325 monoxide-and-ozone-calibration-laboratory-
301 2017. The 43i SO2 analyser is designated by 326 cozi/).
302 the Environmental Protection Agency (EPA) for
303 measurements in the range of 0 - 1000 ppb,
304 with a precision of 1 ppb and a lower detectable
305 SO2 limit of 0.5 ppb [Thermo Scientific 2010;

Page 5
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

Table 2: Frequency of AQMesh pods being offline due to sensor failure or corrosion of key components impacting
ability to function.
AQMesh Installation Measurement Time in Field Number of Number of Measurement
Number Date Site (Days) Occurrences Days Offline Time Offline
1733150 25/02/17 El Panama 187 5 41 22%
712150 01/03/17 El Crucero 183 4 66 36%
789150 27/02/17 Pacaya 184 3 56 30%
803150 27/02/17 Rigoberto 170 3 19 11%
735150 28/02/17 San Juan 157 5 35 22%

Figure 3: Issues with AQMesh pod corrosion and maintenance. (A) corrosion of metal installation mount-
ing bracket is more advanced in measurement sites more frequently impacted by the volcanic plume; (B)
corrosion as a result of the volcanic plume is fast-acting. AQMesh pod 1733150 installed at El Panama
with new metal fittings, padlock and chain (B1) shows obvious signs of corrosion after 14 days (B2); (C)
mounting board for the electrochemical sensors with signs of corrosion on the electrode pins for one of the
sensors; (D) an AQMesh internal computer board which controls the sensors showing signs of corrosion
with one of the board battery units disconnected.

327 2.3 Determination of Plume Direction 348 height of 104 m asl) to 1 hPa (average geopotential
349 height of 46,197 m asl). Measurements of tem-
328 Meteorological observations from ground-based 350 perature (K), wind direction (° from north), wind
329 stations could not be obtained within the scope 351 speed (m/s), geopotential height (m), vertical ve-
330 of this project. Instead we used modelled me- 352 locity (m/s) and humidity (% RH) were extracted
331 teorological data (ECMWF) and high-resolution 353 for each pressure level. The wind speed and direc-
332 satellite imagery (Landlook Viewer) to determine 354 tion were graphically examined in wind roses to
333 plume direction. 355 determine changes in the wind direction through
356 a vertical profile of the atmosphere. The wind di-
334 2.3.1 ECMWF Modelled Data 357 rection was found to be maintained as predom-
358 inantly coming from an East or East-North-East
335 Modelled meteorological data was obtained from 359 direction for the lower 3000 metres of the atmo-
336 the European Centre for Medium-Range Weather 360 spheric column during the data period.
337 Forecasts (ECMWF), which uses ensemble fore- 361 The wind direction at each pressure level from
338 casting to predict the evolution of atmospheric 362 the ECMWF modelled data was compared to the
339 conditions through time [Molteni et al., 1996; 363 wind direction visually derived from satellite im-
340 Buizza et al., 2005]. 364 agery [Section 2.3.2] to determine best correlation
341 Forecast measurements were extracted from the 365 and likely height of the plume. The time of satel-
342 ECMWF model for the grid point at Masaya vol- 366 lite imagery acquisition was obtained from the
343 cano for the period 27th February to 28st Au- 367 image metadata, and the comparison to ECMWF
344 gust 2017. Results were obtained in a three-hour 368 data was calculated at the closest model output
345 cycle, with each output producing modelled re- 369 time to reduce bias from altering wind direction
346 sults at twenty-five pressure levels through the at- 370 before and after satellite imagery capture. The
347 mosphere, from 1000 hPa (average geopotential 371 ECMWF data at 900 hPa had the best correla-

Page 6
372 tion to satellite imagery over the thirteen days 428 plume width, which is not determinable from the
373 of overlapping data, with a Pearson’s r value of 429 ECMWF wind direction data and may cause er-
374 0.86 [Figure 4]. As a result we interpret that 430 ror in the number of AQMesh measurement sites
375 the plume is most commonly at a geopotential 431 which are suggested to have the presence of vol-
376 height of 948.2 m ± 112.9 m asl. The results from 432 canic plume on a specific day.
377 the ECMWF model at 900 hPa were then used
378 throughout the analysis with the AQMesh data.
379 ECMWF data was used in 24-hour averages to give 433 3 Data Analysis
380 a good representation of the overall meteorologi-
381 cal conditions. The frequency of plume dispersal 434 3.1 PM correction factor to remove effect of hu-
382 over each AQMesh measurement site was evalu- 435 midity
383 ated and compared to the AQMesh measurements
384 to interpret the effectiveness of the AQMesh pods 436 The optical particle counter (OPC) used in the
385 at recognising the presence of volcanic plume. 437 AQMesh pods is a small, low-cost sensor mak-
438 ing it suitable for deployment in compact instru-
439 ment systems. Such sensors are becoming widely
386 2.3.2 Satellite Imagery 440 used in the air quality aerosol-monitoring com-
441 munity as they offer an alternative to the expen-
387 Visual satellite imagery was obtained for 442 sive reference-grade instrumentation which often
388 the region around Masaya volcano from 443 require high power input and surrounding infras-
389 the USGS Landlook Viewer (available at 444 tructure [Sousan et al., 2016a; Lewis et al., 2016;
390 https://landlook.usgs.gov/). The Landlook 445 Sousan et al., 2016b; Kelly et al., 2017]. How-
391 Viewer displays high-resolution satellite images 446 ever, the trade-off of using these low-cost compact
392 from Sentinel 2, Landsat 7 and Landsat 8. The 447 OPCs is that they do not currently provide such
393 satellite imagery obtained was non-continuous 448 precise, accurate or sensitive measurements as
394 over 2017. It did, however, allow an opportunity 449 their reference-grade counterparts [Sousan et al.,
395 to "truth" the accuracy of the modelled ECMWF 450 2016a; Crilley et al., 2018, 2020].
396 data, providing a visual check of meteorolog- 451 Part of the issue lies with the methodology for
397 ical conditions and direction of plume travel. 452 acquiring the measurements of the number and
398 Twenty-seven satellite images were obtained of 453 size of the particles. Many low-cost OPCs measure
399 the region around Masaya volcano between 5th 454 the number of particles and the particle diame-
400 March and 28th August 2017. Of these, twelve 455 ters by examining the light-scattering as each par-
401 could not be effectively examined to determine 456 ticle passes through a laser. These measurements
402 plume direction due to the extent of opaque 457 are then converted to particle mass concentrations
403 cloud cover, and two occurred during periods 458 by assuming that the particles are spherical and
404 of unavailable ECMWF data. The remaining 459 of a uniform density. However, most low-cost
405 thirteen satellite images were compared to the 460 OPCs do not dry the sampled air prior to mea-
406 ECMWF database of wind directions. Visual 461 surement, as this would require additional hard-
407 inspection of the plume trajectory was also used 462 ware and power costs. The resulting issues with
408 to determine which AQMesh station was likely 463 measurement accuracy occur most significantly in
409 exposed to volcanic plume on a given day. 464 high humidity conditions [Crilley et al., 2018; Ja-
410 Visual analysis of the satellite imagery sug- 465 yaratne et al., 2018; Crilley et al., 2020]. As a con-
411 gested sources of potential bias in the identifica- 466 sequence of this, the reported particle mass con-
412 tion of plume presence at a measurement station 467 centrations from low-cost OPCs need to be con-
413 using modelled meteorological data from the grid 468 verted from wet particle mass concentrations to
414 point of the volcanic source. Some satellite im- 469 dry particle mass concentrations in order to be
415 agery indicated movement of Masaya’s volcanic 470 more accurate and comparable to reference-grade
416 plume in a non-linear direction. A non-linear 471 instruments.
417 plume trajectory (Figure 5A) may result in bias 472 Here we follow the methodology described in
418 in the assignment of which AQMesh measure- 473 Crilley et al. [2018] and Crilley et al. [2020] to
419 ment site was exposed to volcanic plume, as the 474 apply a correction factor to the reported results
420 wind direction does not remain constant follow- 475 for PM1 , PM2.5 and PM10 from the AQMesh net-
421 ing plume dispersal away from the source-point. 476 work in Nicaragua. The correction factor (C) is
422 As there is limited availability of satellite imagery 477 applied in the following manner as described in
423 for use in comparison to the ECMWF data, it is 478 Equation 1;
424 not possible to determine the frequency of occur-
425 rences of non-linear plume trajectory. Another κ
ρp
426 potential source of bias is the width of the plume. C = 1+ (1)
427 Figure 5A and B indicate the variability in lateral −1 + a1
w

Page 7
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

Figure 4: Meteorological data for Masaya volcano. (A) Wind rose for the period February to August 2017 at
900 hPa, with an average geopotential height of 948 m ± 113 m. Wind direction is predominantly from the
ENE and E. Data is derived from ECMWF modelled meteorological data and is displayed as the direction
the wind is blowing from. (B) Comparison between the ECMWF 900 hPA wind direction data and the wind
directions derived from the satellite imagery over Masaya volcano.

479 where ρp is the density of the dry particles (here 511 reference-grade instruments for pollution moni-
480 we use 1.65 g cm-3 which is the ambient particle 512 toring [Lewis et al., 2016]. Electrochemical sen-
481 density assumed by the OPC-N2); aw is the wa- 513 sors operate by diffusing the target gas through
482 ter activity (RH/100) and the value for κ can be 514 a porous membrane, following which changes in
483 found by a non-linear curve fitting of a humido- 515 the chemical potential are measured by a sensing
484 gram (aw vs m/m0 where m and m0 are the wet 516 electrode [Austin et al., 2006; Mead et al., 2013].
485 and dry (RH=0%) aerosol mass, respectively). For 517 However, other substances may interfere with the
486 PM1 and PM2.5 measurements we used a κ value 518 chemical potential of the electrochemical sensor,
487 of 0.53 relating to ammonium sulphate and for 519 causing a positive or negative interference to the
488 PM10 measurements we use a κ value of 0.33 re- 520 sensor output, resulting in a biased measurement
489 lating to dust particles. For the correction factor 521 [Austin et al., 2006; Lewis et al., 2016]. These in-
490 calculations we use the relative humidity as mea- 522 terfering substances can include a number of com-
491 sured by the humidity sensor installed in the base 523 pounds, only some of which are reported by the
492 plate of the AQMesh pods. 524 manufacturer. Studies such as Mead et al. [2013]
493 The raw particle mass concentrations reported 525 have shown that electrochemical sensors are suit-
494 by the AQMesh pods can then be corrected ac- 526 able for monitoring gas levels in low ppb concen-
495 cording to Equation 2: 527 trations, but it must also be recognised that cross-
528 sensitivities of the sensors may have a significant
P M Raw 529 impact on the sensor output.
P M Corr = (2)
C 530 The cross-sensitivites of the SO2 -
496 During the measurement period of the AQMesh 531 B4 sensor used in this study can be
497 pods in Nicaragua, there was no period of cali- 532 found in the instrument data sheet
498 bration with a reference-grade particulate instru- 533 (https://www.alphasense.com/WEB1213/wp-
499 ment for the OPC-N2s within the AQMesh pods. 534 content/uploads/2019/09/SO2-B4.pdf). Of
500 However, the manufacturer states that the OPC- 535 these, cross-sensitivities to NO2 and O3 are
501 N2 instruments are factory calibrated prior to 536 the most likely to impact the accuracy of the
502 sale, and the application of the correction fac- 537 SO2 measurement. Changes in temperature
503 tor as described above should remove the impact 538 and humidity can also impact electrochemical
504 of high humidity conditions in the field [Crilley 539 sensor performance [Mead et al., 2013; Lewis
505 et al., 2018, 2020]. 540 et al., 2016]. However, in this study we are not
541 investigating quantitative concentrations of SO2
542 so much as the AQMesh pods’ efficiency at de-
506 3.2 Electrochemical Sensor Cross-Sensitivities
543 termining relative correlations between SO2 and
507 Low-cost air pollution sensors, such as the elec- 544 particulates and hence the equipment’s ability
508 trochemical gas sensors used in this study, are 545 to recognise the presence or absence of volcanic
509 generally not as accurate or precise at measuring 546 plume, and as such the absolute concentrations of
510 concentrations of pollutants in the atmosphere as 547 SO2 are not required.

Page 8
Figure 5: Satellite imagery obtained from the USGS Landlook viewer, annotated with plume trajectories
indicated by white dashed lines. The source point (Masaya volcano’s Santiago crater) is visible with the
lava lake. AQMesh measurement stations are indicated with black circles and labelled as follows: ElC - El
Crucero; P - Pacaya; ElP - El Panama; R - Rigoberto; SJ - San Juan. Plume is visible by semi-linear feature of
white condensing clouds initiating from the source point, often interspersed with blue-tinged haze which
is likely due to the particulate component. (A) 13th March 2017 where the plume moves initially towards
the south-west before the trajectory alters towards the west. Plume width is approximately 1.5 km. (B)
30th March 2017 where the plume moves west with a wide lateral spread of approximately 3 km within
the first 4 km from the source point.

548 3.3 Data Processing 575 of the electrochemical SO2 sensor. At Rigoberto
576 station, the AQMesh pod recorded some instances
549 The data obtained from the AQMesh instruments 577 where SO2 concentrations peaked at 300 ppb for
550 was processed into hourly averages and graphed 578 short (< 3 hour) durations. Such peaks in SO2
551 for initial visual analysis. Periods of erroneous or 579 were not found elsewhere during the analysis and
552 missing data (Table 2) were not included in the 580 have been interpreted to be anthropogenic pollu-
553 analysis. The remaining data was analysed us- 581 tion, likely from the cooking fire which was lo-
554 ing RatioCalc 3.2 following Tamburello [2015] in 582 cated close by. These cooking fire related correla-
555 order to determine when there were correlations 583 tions were removed from the analysis.
556 between the gas and particulates, indicating the 584 A further analysis was done on the PM1 , PM2.5
557 presence of volcanic plume. RatioCalc was used 585 and PM10 data from each of the measurement
558 to determine periods of time, greater than four 586 sites. Using the plume-dispersal identification
559 days in duration, when there were strong corre- 587 from ECMWF and satellite imagery, the particu-
560 lations between SO2 and the corrected measure- 588 late data was divided into categories of plume-
561 ments of PM1 , PM2.5 and PM10 , respectively. A 589 present and plume-absent conditions and anal-
562 minimum of four days of correlation was set to 590 ysed to investigate any significant differences in
563 restrict correlations based on limited data-points. 591 the number and range of particulates under the
564 Only correlations between SO2 and the PM size 592 different conditions. A two-sample t-test was
565 fractions which had an r2 value greater than 0.5 593 applied to determine the significance level of
566 were included. Data were excluded from the anal- 594 any differences between the plume-present and
567 ysis where the SO2 had a correlation to the rel- 595 plume-absent conditions for each of the size frac-
568 ative humidity greater than an r2 value of 0.25, 596 tion categories. Using the plume-absent data-
569 as a dependence on humidity indicates the SO2 597 set, a monthly average background concentra-
570 measurements are not representative of the actual 598 tion was calculated for each particulate size cat-
571 atmospheric SO2 concentrations. Correlation pe- 599 egory at each measurement station. The plume-
572 riods were only included where SO2 concentra- 600 present and plume-absent data-sets where then
573 tion had peaks above 20 ppb, as concentrations 601 cross-referenced against the monthly background
574 lower than this were within the accuracy range 602 average to determine significant outliers indicat-

Page 9
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

603 ing a higher concentration of particulates than 657 between the electrochemical sensors and the SO2
604 would be expected under background conditions. 658 analyser was very good, with Pearson’s r values of
659 0.92 and 0.93 respectively across the two AQMesh
605 4 Results 660 pods. The two AQMesh pods correlated extremely
661 well to each other, with a Pearson’s r value of 0.98
606 4.1 Co-Location of AQMesh Pods 662 and only 5% variance in the measurement trend
663 (Figure 7C). AQMesh pods may not be suitable
607 The AQMesh pods were co-located in a measure- 664 for reporting absolute values of SO2 , but are re-
608 ment site away from volcanic input for eleven 665 liable for measuring when SO2 in the atmosphere
609 days [section 2.2]. During this time the SO2 con- 666 is increased to above background concentrations.
610 centrations were low and within the lower noise 667 We therefore suggest that they are suitable for this
611 limit for the electrochemical sensors. As such, 668 low-cost sensor network where the instruments
612 only the particulate measurements from this co- 669 are used to determine the presence or absence of
613 location period were considered, the results of 670 volcanic plume and where the absolute values of
614 which for PM1 and PM2.5 are indicated in Fig- 671 SO2 concentration are not required.
615 ure 6. 672 Although AQMesh indicates that the SO2 sen-
616 The particulate measurements of each of the 673 sors were calibrated in the factory prior to sale, it
617 five AQMesh pods were analysed and assessed 674 is likely that each electrochemical sensor will have
618 for correlation during the co-location period. For 675 a slightly different level of accuracy and measure-
619 PM1 measurements, Pearson’s r values were high 676 ment precision with the potential for some base-
620 between instruments, with 30% of the r values 677 line drift. The impact of the corrosive volcanic
621 above 0.9 and 60% above 0.8. However there was 678 environment meant that SO2 electrochemical sen-
622 some significant variability in the measurements 679 sors consistently failed in the field and required
623 recorded, with AQMesh pods 712150 and 789150 680 replacing. This means that there will be varying
624 consistently measuring higher values than pods 681 levels of confidence in the AQMesh SO2 measure-
625 803150, 1733150 and 703150. For PM2.5 mea- 682 ments over the course of the campaign, adding an
626 surements, similar Pearson’s r values were found, 683 additional level of complication to the data analy-
627 with 20% above 0.9 and 70% above 0.8. Mea- 684 sis.
628 surement variability was reduced in PM2.5 mea-
629 surements, though pod 712150 consistently re-
685 4.3 Efficiency of AQMesh pods recognising vol-
630 ported lower values with respect to the other in-
686 canic plume
631 struments.
632 The co-location indicated that there was some 687 4.3.1 Correlations between SO2 and particulates
633 significant variability between the OPC-N2 in-
634 struments within the AQMesh pods at concen- 688 The ECMWF data at 900 hPa pressure and the
635 trations below 15 µg/m3 for PM1 and below 689 satellite imagery were analysed to determine
636 30 µg/m3 for PM2.5 . However, despite the vari- 690 when the volcanic plume was likely to be present
637 ability in absolute measurements, all the in- 691 at an AQMesh measurement site. This database
638 struments recorded high concentrations simul- 692 was then cross-referenced with the AQMesh cor-
639 taneously during periods of increased atmo- 693 relation periods to determine the ability of the
640 spheric particles, suggesting that the AQMesh 694 instruments to detect the presence of the plume.
641 pods are suitable for monitoring increases in PM 695 The results are presented in Figure 8.
642 above background concentrations, even if there 696 No suitable correlations between SO2 and the
643 is variability in the absolute concentration of PM 697 corrected PM measurements were found at the
644 recorded. 698 AQMesh measurement sites of Pacaya and San
699 Juan. As Table 2 indicates, Pacaya station had a
700 relatively high frequency of maintenance issues
645 4.2 Co-Location of SO2 Analyser and AQMesh
701 resulting in limited data, none of which indicated
646 Pods
702 the presence of volcanic plume. The AQMesh
647 The pulsed fluorescence spectroscopy SO2 anal- 703 pod at San Juan also failed to indicate the pres-
648 yser was co-located with two AQMesh instru- 704 ence of volcanic plume, with no suitable correla-
649 ments for two days to determine the accuracy of 705 tions found between SO2 and PM measurements.
650 the SO2 electrochemical AQMesh sensors in the 706 San Juan is located on the edge of the prevail-
651 field after several months of deployment [Sec- 707 ing plume trajectory area [Figure 2] and as such
652 tion 2.2]. There was good agreement between the 708 may have been exposed to volcanic plume rarely
653 two instrument types, though both AQMesh elec- 709 or for short durations during the measurement
654 trochemical sensors underestimated the concen- 710 period. Analysis of the ECMWF data and satel-
655 trations measured by the SO2 analyser by up to 711 lite imagery indicate there were no periods when
656 75%, as indicated in Figure 7A, B. The correlation 712 the volcanic plume was dispersed towards the San

Page 10
Figure 6: Scatter plot matrix of PM1 and PM2.5 results from co-location of AQMesh pods for eleven days in
July 2017. PM1 results are displayed in the ten plots in the upper triangle with data plotted in blue. PM2.5
results are displayed in the ten plots in the lower triangle with data plotted in purple. All data presented
have been processed with the correction factor outlined in section 3.1.

713 Juan AQMesh station during the measurement pe- 733 of AQMesh pods identifying plume presence.
714 riod. 734 The meteorological data were analysed to form
715 The El Panama AQMesh pod was the most ef- 735 a database of the total frequency of interpreted
716 fective at recognising the presence of the plume. 736 plume presence over each individual measure-
717 At El Panama the AQMesh pod positively iden- 737 ment station, combining both ECMWF and satel-
718 tified the plume with SO2 to PM2.5 correlations 738 lite plume events. The AQMesh correlation
719 65% of the time for ECMWF-derived plume pres- 739 periods were cross-analysed to determine to-
720 ence and 42% of the time for plume presence as 740 tal frequency of true-positive identification of
721 derived from satellite imagery [Figure 8]. Rigob- 741 plume presence by the AQMesh instruments. A
722 erto station correctly identified the presence of the 742 null hypothesis test was then implemented by
723 plume by correlations of SO2 to both PM1 and 743 analysing the meteorological data to determine
724 PM2.5 on 42% of ECMWF-derived plume events 744 from ECMWF and satellite imagery the frequency
725 and 30% of satellite-derived plume events [Fig- 745 of volcanic plume being dispersed to areas other
726 ure 8]. El Crucero AQMesh pod only recognis- 746 than a specific measurement site. During these
727 ing a volcanic signature only 5% of the time that 747 periods when the plume was predicted to be
728 the ECMWF data indicated the plume was present 748 dispersed away from the measurement site, the
729 over the station, and 0% of the time for satellite- 749 AQMesh data was analysed to check for false-
730 derived plume presence [Figure 8]. 750 positive recognition of the plume’s presence. The
751 results are presented in Figure 9.
731 A control analysis was implemented to iden-
732 tify the ratio of true-positives and false-positives 752 At El Panama station the AQMesh pod recorded

Page 11
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

Figure 7: Co-location of 43i SO2 analyser and two AQMesh pods at El Crucero measurement site. Com-
parisons of SO2 measurements between (A) 43i analyser and the SO2 electrochemical sensor in the 712150
AQMesh pod; (B) 43i analyser and the SO2 electrochemical sensor in the 703150 AQMesh pod; (C) SO2
electrochemical sensor in the 712150 AQMesh pod and the SO2 electrochemical sensor in the 703150
AQMesh pod. Comparisons and data presented here are the peaks in data, near-baseline measurements
were removed as they were within the baseline noise fluctuations of the AQMesh sensors. Measurements
from the 43i analyser have been corrected to remove baseline drift and to account for 18% underestimation
as indicated by post-fieldtrip calibration.

753 a high success rate of true-positive identifications 786 with a significance level of 0.05) and at El Panama
754 of the presence of volcanic plume, with correla- 787 (probability 6.3 x 10-10 for PM1 , 9.6 x 10-8 for
755 tions of SO2 to PM. Correlations indicating the 788 PM2.5 and 5.9 x 10-7 for PM2.5 with a significance
756 presence of the volcanic plume provided true- 789 level of 0.05). The two-sample t-test at Pacaya
757 positive identifications 59% of the time for SO2 790 station showed a lower level of significance in
758 and PM1 , 64% of the time for SO2 and PM2.5 , 791 the difference between plume-present and plume-
759 and 46% of the time for SO2 and PM10 [Fig- 792 absent conditions (probability 5.4 x 10-3 for PM1 ,
760 ure 9]. There were significantly lower num- 793 2.7 x 10-4 for PM2.5 and 3.9 x 10-2 for PM2.5 with a
761 bers of false-positive plume identifications at El 794 significance level of 0.05). Rigoberto data proved
762 Panama, at 18% for PM1 and PM10 and 23% for 795 to have a non-significant difference between the
763 PM2.5 [Figure 9]. Rigoberto AQMesh pod likewise 796 plume-present and -absent conditions (probabil-
764 had a higher percentage of true-positive instances 797 ity 0.73 for PM1 , 0.88 for PM2.5 and 0.58 for PM2.5
765 than false-positives for correlations between SO2 798 with a significance level of 0.05).
766 and both PM1 (41% true-positive and 22% false- 799 The maximum PM concentrations for each size
767 positive) and PM2.5 (41% true-positive and 15% 800 fraction do not appear to be linked to the presence
768 false-positive), and with no correlations found be- 801 or absence of the plume, with equally high con-
769 tween SO2 and PM10 in either instance. How- 802 centrations recorded under both plume-present
770 ever, the absolute number of false-positive identi- 803 and plume-absent conditions [Figure 10]. How-
771 fications was higher at Rigoberto for SO2 to PM1 , 804 ever, analysis of the data shows that the av-
772 with 16 true-positive events and 21 false-positive 805 erage PM concentration for all size fractions is
773 events. El Crucero AQMesh station had an equal 806 higher when the plume is present at the measure-
774 percentage of true- and false-positive events for 807 ment station than when the plume is absent [Fig-
775 SO2 to PM1 correlations, and no correlations be- 808 ure 10]. From this we infer that there is a "back-
776 tween SO2 and either PM2.5 or PM10 . 809 ground" level of particulates which is present un-
810 der both plume-absent and -present conditions,
777 4.3.2 Particulate Analysis 811 and in plume-present conditions there is an addi-
812 tional volume of particles which are volcanically-
778 An analysis of particulate concentrations was 813 sourced.
779 done to determine whether there were signifi- 814 Daily average particulate concentrations for
780 cant differences in the number and sizes of par- 815 PM1 , PM2.5 and PM10 were calculated for each
781 ticulates under plume-present and plume-absent 816 measurement station. The plume-absent data was
782 conditions. A two-sample t-test indicated that 817 used to create a monthly average of "background"
783 plume-present and -absent conditions varied sig- 818 particulate concentrations, which was then com-
784 nificantly at El Crucero (probability 2.6 x 10-27 for 819 pared to the daily average data for both plume-
785 PM1 , 1.5 x 10-14 for PM2.5 and 1.6 x 10-4 for PM2.5 820 present and plume-absent conditions. We iden-

Page 12
Figure 8: Efficiency of AQMesh pods at recog- Figure 9: Efficiency of AQMesh pods at recog-
nising the presence of volcanic plume derived nising whether the volcanic plume was present
from satellite and ECMWF data at three mea- at three measurement sites. Black bars indicate
surement sites. The frequency of plume pres- frequency of plume presence at the measurement
ence at the measurement site is indicated by the site, as derived from meteorological data, dur-
black bar, and frequency of correlation between ing periods when the AQMesh pod was fully-
AQMesh SO2 and PM measurements are indicated functional. Yellow hatched bars indicate fre-
by the blue, green and pink bars. Only periods of quency of no volcanic plume at the measurement
AQMesh instrument functionality are considered site, as derived from meteorological data, dur-
here. Results are split into frequency of plume ing periods when the AQMesh pod was fully-
presence derived from ECMWF database and de- functional. Percentages noted on each coloured
rived from satellite imagery. Percentages noted bar under "plume present" conditions indicate
on each coloured bar indicate the proportion of the proportion of frequency that the AQMesh
frequency that the AQMesh pods recognised the pods recognised the presence of plume. Percent-
presence of plume derived from the relevant me- ages noted on each coloured bar under "plume
teorological data. not present" conditions indicate the proportion
of frequency that the AQMesh pods gave a false-
positive and falsely indicated the presence of
plume.

Page 13
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

Figure 10: Range of particulate concentrations at El Crucero, El Panama and Rigoberto measurement
stations under plume-present (brown box-plots) and plume-absent (blue box-plots) conditions. Plume-
presence or -absence is as identified by ECWMF data and satellite imagery. Note the logarithmic y-axis
scale.

821 tified events where the average daily concentra- 840 plume-present and -absent conditions [Table 3].
822 tion exceeded one standard deviation above the
823 background average, indicating a pollution event
824 which was significantly above the background 841 5 Discussion
825 norm. At El Panama there was a much higher fre-
826 quency of events exceeding one standard devia- 842 5.1 AQMesh Pods for Monotoring Volcanic En-
827 tion above the background average during peri- 843 vironments
828 ods when the plume was present (40.7% for PM1 ,
844 5.1.1 Correlations between SO2 and particulates
829 39.5% for PM2.5 and 44.4% for PM10 ) than when
830 the plume was absent (16.5% for PM1 , and 21.2% 845 The AQMesh pods at both El Panama and Rigob-
831 for both PM2.5 and PM10 ) [Table 3]. Similarly El 846 erto were reasonably effective at recognising and
832 Crucero measurement station had a higher fre- 847 reporting the presence of volcanic plume in the
833 quency of exceedances above one standard devi- 848 form of correlations between SO2 and PM dur-
834 ation of the background average during plume- 849 ing periods when the meteorological data sug-
835 present conditions (47.5% for both PM1 and PM2.5 850 gested the plume was over the measurement sta-
836 and 25% for PM10 ) than when the plume was ab- 851 tion. Both stations also had a number of false-
837 sent (8.6% for PM1 , and 5.5% for PM2.5 and 12.6% 852 positive plume identifications, though the per-
838 for PM10 ). However, both Pacaya and Rigob- 853 centage of false-negatives was lower in all cases
839 erto stations had much smaller variations between 854 than the percentage of true-positives [Figure 9].

Page 14
Table 3: Frequency of events exceeding 1 standard deviation (SD) above monthly background average particulate
concentration. Plume-present and plume-absent conditions are determined from ECMWF data and satellite imagery,
and the monthly background average concentration is calculated from the plume-absent data.
Events exceeding 1∼SD Time exceeding 1∼SD above
Station Particulate
above background average background average
Location Size
Plume-Present Plume-Absent Plume-Present Plume-Absent
Group
Conditions Conditions Conditions Conditions
PM1 33 14 40.7% 16.5%
El Panama PM2.5 32 18 39.5% 21.2%
PM10 36 18 44.4% 21.2%
PM1 19 11 47.5% 8.6%
El Crucero PM2.5 19 7 47.5% 5.5%
PM10 10 16 25.0% 12.6%
PM1 16 13 22.2% 17.6%
Pacaya PM2.5 18 15 25.0% 20.3%
PM10 13 14 18.0% 18.9%
PM1 9 20 21.4% 16.3%
Rigoberto PM2.5 6 20 14.3% 16.3%
PM10 6 18 14.3% 14.6%

855 These findings indicate that the AQMesh pods 890 correlation between SO2 and particulates in the
856 are reasonably suitable for recognising the pres- 891 size fractions of PM1 , PM2.5 and PM10 , respec-
857 ence of a volcanic plume by means of correlations 892 tively. The particulates in this instance will likely
858 between SO2 and PM in near-field measurement 893 be sulphate aerosol which is a dominant compo-
859 sites. 894 nent of volcanic plumes [Pattantyus et al., 2018;
895 Tam et al., 2016]. Sulphate aerosol is emitted
860 The AQMesh pods at El Crucero and Pacaya
896 from the volcanic source as a primary aerosol,
861 measurement sites were the least efficient at
897 and will also exist in the volcanic plume as sec-
862 recognising the presence of the volcanic plume,
898 ondary aerosol forming from the conversion of
863 with only minimal volcanic correlations recorded
899 SO2 in the atmosphere [Stockwell and Calvert,
864 at El Crucero and none at Pacaya station. This in-
900 1983; Allen et al., 2002]. The relatively near-
865 efficiency at recognising plume presence may be
901 source measurement stations of El Panama and
866 due to specific conditions in the localised envi-
902 Rigoberto will likely be exposed to volcanic plume
867 ronment of the monitoring stations that biased the
903 rich in both primary and secondary aerosol as well
868 effectiveness of the AQMesh pods. Alternatively,
904 as SO2 . However, the measurement stations at El
869 the inefficiency of Pacaya and El Crucero pods
905 Crucero and Pacaya are significantly further from
870 at recognising the presence of the plume may be
906 the source point and, as such, the volcanic plume
871 related to the distance between the instrument
907 will have chemically matured to be dominantly
872 site and the source point. El Crucero and Pacaya
908 composed of secondary aerosol with a lower con-
873 measurement stations are situated approximately
909 centration of SO2 following the atmospheric con-
874 16 km from Santiago Crater, as opposed to 4 km
910 version process. With this being the case, analy-
875 for both El Panama and Rigoberto. This greater
911 sis for correlations between SO2 and particulates
876 distance may result in significant divergence of
912 in the data collected at El Crucero and Pacaya
877 the plume dispersal direction. The ECMWF data
913 measurement stations are likely to produce fewer
878 was extracted for the grid point local to Santi-
914 results, particularly taking into account that we
879 ago Crater and may not be representative of wind
915 use only correlations in the data where SO2 con-
880 conditions over the distance between the volcanic
916 centrations spike above 20 ppb. For this reason,
881 source point and the Las Sierras highlands. Error
917 we also analyse the particulate concentrations at
882 in the meteorological data may also be introduced
918 all of the measurement sites in order to establish
883 by non-linear dispersion of the volcanic plume, as
919 whether low-cost particulate monitoring at far-
884 indicated in Figure 5A. As such, the apparent in-
920 field sites, such as El Crucero and Pacaya, can ef-
885 efficiency of the El Crucero and Pacaya AQMesh
921 fectively identify the presence of volcanic plumes.
886 pods may be because the plume was not physically
887 present at the stations at times when the ECMWF
922 At the three stations which were found to have
888 data and satellite imagery suggests it would be.
923 correlations between SO2 and PM, the AQMesh
889 The volcanic signature we analyse for here is a 924 pods were more efficient at recognising the pres-

Page 15
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

925 ence of plume in instances when the plume pres- 981 knowledge, there are no official records kept re-
926 ence was derived from the ECMWF data as op- 982 lating to anthropogenic fires or wild fires, so val-
927 posed to derived from satellite imagery [Figure 8]. 983 idation of this hypothesis cannot be tested. Al-
928 This is potentially due to satellite images provid- 984 ternatively, the above-background events during
929 ing only a snapshot of the meteorological condi- 985 plume-absent conditions may be related to bias or
930 tions, whereas the ECMWF data were analysed as 986 errors in the identification of plume location from
931 daily averages, providing an overview of condi- 987 the ECMWF data or satellite imagery.
932 tions and reducing the impact of outlying wind 988 The AQMesh pod at Rigoberto measurement
933 directions. 989 station had a non-significant finding for all par-
934 The issue with using meteorological data, per- 990 ticulate size fractions between plume-present and
935 haps especially modelled meteorological data and 991 plume-absent conditions in the two-sample t-test.
936 satellite imagery, is that we cannot be certain of 992 The particulate analysis also indicated a low vari-
937 the presence or absence of the volcanic plume at 993 ability in the number of above-background events
938 a measurement station on any given day. Addi- 994 between plume-present and plume-absent condi-
939 tionally, although meteorological data may cor- 995 tions. We consider that these findings may be
940 rectly suggest the plume is being dispersed to- 996 due to the reasonably close proximity of Rigoberto
941 wards a measurement station on a given day, there 997 measurement station to domestic cooking fires.
942 is no guarantee that the plume will "ground" and 998 Pacaya measurement station likewise had a low
943 be measurable at ground-level. Were the exper- 999 variability in the number of above-background
944 iment to be repeated, it would be invaluable to 1000 events between plume-present and plume-absent
945 set up a reference-grade measurement station in 1001 conditions. The cause for this low variability is
946 one of the AQMesh sites to verify the presence or 1002 undetermined, but may be as a result of localised
947 absence of volcanic plume and test the response 1003 anthropogenic pollution sources.
948 of the AQMesh pod relative to the reference- 1004 Overall the AQMesh pods in proximal locations
949 grade instrument. This would allow verification 1005 (~ 4 km) from the volcanic source point provided
950 of whether the AQMesh pods were correct in their 1006 reasonable results of identifying the presence of
951 positive or negative identification of the plume’s 1007 the volcanic plume using the method of identify-
952 presence. 1008 ing correlations between SO2 and PM size frac-
1009 tions. However, the far-field measurement sites
1010 at El Crucero and Pacaya (~ 16 km from the vol-
953 5.1.2 Particulate Analysis
1011 canic source point) were inefficient at recognising
954 At El Panama and El Crucero, the two-sample t- 1012 a volcanic signature using SO2 to PM correlations.
955 test indicated a higher significance in the differ- 1013 However, when investigating the particulate data
956 ence between plume-present and -absent condi- 1014 from all measurement stations, examples of both
957 tions for PM1 than for the larger size fractions. 1015 near-field (El Panama) and far-field (El Crucero)
958 This indicates that the majority of the volcanic 1016 measurement sites were proved to be reasonably
959 particulates are small (<1 µm diameter), which 1017 effective at recognising the presence of the vol-
960 follows the findings of previous studies [Ilyin- 1018 canic plume. However with particulate analysis it
961 skaya et al., 2017, 2021; Mason et al., 2021] where 1019 appears very important what other potential pol-
962 volcanic particulates are often found to be in the 1020 lution sources are close by, as indicated by the
963 smallest size fraction. 1021 AQMesh pod at Rigoberto measurement site. As
964 The findings from the particulate analysis from 1022 such, the AQMesh pods are reasonably suitable
965 the AQMesh pods indicate that the low-cost net- 1023 for monitoring the presence of volcanic plume.
966 work is reasonably suited for monitoring the pres-
967 ence or absence of volcanic plume using the PM 1024 5.2 Alternative Monitoring Options
968 concentrations, if there is a suitable sampling pe-
969 riod without the presence of volcanic plume in 1025 The AQMesh pods provide an opportunity for a
970 order to determine background concentrations. 1026 relatively low-cost monitoring network in a region
971 At both El Panama and El Crucero measurement 1027 which otherwise has no means for long-term mea-
972 stations, the AQMesh pods recorded significantly 1028 surements of volcanic plume concentrations in the
973 higher frequencies of above-background particu- 1029 downwind communities. The instruments cost in
974 late concentrations during plume-present condi- 1030 the range of £7000 per pod with a yearly sub-
975 tions than plume-absent conditions [Table 3]. We 1031 scription of £480 for use of the online web server.
976 infer that the above-background concentrations 1032 A network of five AQMesh instruments for per-
977 during plume-absent conditions may be related to 1033 manent installation would therefore have an ini-
978 anthropogenic pollution events such as cooking 1034 tial cost of £35,000, followed by yearly server and
979 fires or agricultural fires, or be related to wild- 1035 maintenance costs. They are easy to install and
980 fire events. However, to the best of the authors’ 1036 do not require a large amount of infrastructure to

Page 16
1037 operate. However, as indicated in this study, they 1095 [PurpleAir, 2019] and each instrument contains
1038 suffer from frequent instrument and system fail- 1096 two Plantower PMS5003 sensors mounted in one
1039 ure in volcanic environments due to the high level 1097 housing, allowing self-consistency checks to alert
1040 of fast-acting corrosion from the volcanic plume 1098 when significant differences are reported between
1041 (Figure 3). In such instances they require con- 1099 the internal sensors. The instruments are easy
1042 tinual maintenance and purchase of replacement 1100 to install, though do require mains power and
1043 parts. Although having the advantage of being 1101 stable wifi access. Although PurpleAirs are un-
1044 low-cost and relatively compact, the AQMesh in- 1102 able to monitor SO2 , their easily accessible open-
1045 struments have the fallback of being less precise 1103 access web server allows a community-operated
1046 and accurate than larger reference-grade instru- 1104 network of instruments to provide information
1047 ments, as indicated in Figure 7. The data from 1105 across a wide spatial area at a cost that is rela-
1048 the instruments also requires significant process- 1106 tively very low in comparison to both reference-
1049 ing and analysis, meaning that the pods are not 1107 grade and AQMesh networks. During the 2018
1050 useful for active plume tracking in real time. 1108 eruption of Kı̄lauea’s lower East Rift Zone on the
1051 In comparison, if a network of reference-grade 1109 Island of Hawai‘i, PurpleAir instruments were in-
1052 monitoring instruments was installed, the cost of 1110 stalled across the island by community members
1053 installation and construction of the required in- 1111 and provided an open-access source of air qual-
1054 frastructure would be significantly higher. A sin- 1112 ity information with a reasonably high level of ac-
1055 gle pulsed fluorescence spectroscopy SO2 analyser 1113 curacy as compared to reference-grade particulate
1056 costs in the range of £8,000 to £12,000 depending 1114 sensors [Whitty et al., 2020].
1057 on the configuration (quote from Thermo Fisher
1058 Scientific, 2021). A BAM-1020 (manufactured 1115 5.3 Recommendations for future use of AQMesh
1059 by Met One Instruments) which is a reference- 1116 pod networks in volcanic environments
1060 grade FEM instrument for measuring PM would
1061 cost in the range of £17,000 (quote from En- 1117 Were the AQMesh pods to be installed as perma-
1062 viro Technology Services Ltd, 2021). The instru- 1118 nent monitoring stations in a volcanic environ-
1063 mentation cost of one reference-grade monitor- 1119 ment like Nicaragua, there are a number of prac-
1064 ing station would therefore be in the range of 1120 tical measures which could be implemented to in-
1065 £25,000 to £29,000, and a network of five would 1121 crease the effectiveness of the instruments. Reg-
1066 be approximately £135,000 for instrument pur- 1122 ular in-field calibration using reference-grade in-
1067 chase. Permanent ambient air quality stations 1123 struments would allow for the accuracy and preci-
1068 such as these must be kept in air-conditioned en- 1124 sion of the low-cost instruments to be assessed, es-
1069 closures to maintain long-term stability, adding 1125 pecially following replacement of electrochemical
1070 significant additional cost and time considera- 1126 sensors. In-field testing of electrochemical cross-
1071 tions related to construction of buildings with ac- 1127 sensitivity would also be advisable.
1072 cess to mains power. Although such a network 1128 In Nicaragua one of the main issues facing the
1073 of reference-grade monitoring instruments would 1129 performance of the AQMesh pods was the fast-
1074 provide accurate and reliable measurements of 1130 acting high level of corrosion resulting from the
1075 volcanic plume in downwind locations, the fea- 1131 volcanic environment. Where feasible this could
1076 sibility of installing such a high-expense moni- 1132 be at least partially mitigated by shielding of
1077 toring network is restricted in locations such as 1133 metal components within the AQMesh pods using
1078 Nicaragua where infrastructure is limited. 1134 polyethylene heat shrink film to externally coat
1079 Perhaps an alternative option for a lower-cost 1135 vulnerable components and minimise exposure to
1080 instrument network would be PurpleAir sen- 1136 corrosion.
1081 sors. PurpleAir (Utah, USA) instruments cost 1137 On-site measurements of humidity are crucial
1082 approximately $250 per unit (quote from Pur- 1138 to allow effective analysis of the particulate mea-
1083 pleAir, 2019) and can be purchased and operated 1139 surements and correction of the results to reduce
1084 by individuals with all data available online in 1140 the impact of high-humidity conditions [Crilley
1085 an open-access, very user-friendly format [Pur- 1141 et al., 2018, 2020]. Humidity measurements are
1086 pleAir, 2019]. PurpleAir instruments only have 1142 also important to allow filtering of the SO2 mea-
1087 the capability to measure particulates, which they 1143 surements by the electrochemical sensors, as high
1088 measure using Plantower PMS5003 nephelometer 1144 correlations between recorded SO2 and humid-
1089 sensors operating on a light-scattering principal 1145 ity indicate that the sensor has a dependence on
1090 similar to the OPC-N2s within an AQMesh pod. 1146 the environmental conditions and is not record-
1091 PurpleAirs have a 10 second response time and 1147 ing true SO2 concentrations. Where funds and in-
1092 can detect particles between 0.3 µm - 10 µm in di- 1148 frastructure allow, installation of a ground-based
1093 ameter [Kelly et al., 2017; Sayahi et al., 2019]. The 1149 meteorological station at the volcanic source point
1094 instruments are factory calibrated prior to sale 1150 would be invaluable to allow identification of

Page 17
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

1151 wind conditions and direction of plume dispersal 1205 ECMWF modelled meteorological data and from
1152 on a day-to-day basis. 1206 a visual inspection of plume dispersal from high-
1207 resolution satellite imagery. The data from the
1153 5.4 Volcanic gas exposure mitigation 1208 AQMesh pods were analysed to determine vol-
1209 canic signature by correlations between SO2 and
1154 Exposure to volcanic gases and aerosols can cause 1210 PM1 , PM2.5 and PM10 , respectively. The partic-
1155 long-term health impacts and result in signifi- 1211 ulate data was also analysed separately to de-
1156 cant issues, particularly for children and vulner- 1212 termine differences in size and concentration of
1157 able individuals including asthmatics [CRI, 2004; 1213 particles under plume-present and plume-absent
1158 ATSDR, 1998]. In the areas surrounding persis- 1214 conditions.
1159 tently degassing volcanoes, like Masaya, expo- 1215 Of the three AQMesh pods which were anal-
1160 sure can occur on a daily basis over years. De- 1216 ysed, El Panama and Rigoberto pods were reason-
1161 pending on the rate of volcanic degassing, the 1217 ably effective at positively identifying the pres-
1162 meteorological conditions and the characteris- 1218 ence of the volcanic plume using correlations be-
1163 tics of the plume, exposure to volcanic gas and 1219 tween SO2 and PM during periods when meteo-
1164 aerosol in any one location may fluctuate with 1220 rological data indicated plume dispersal towards
1165 time. If there is good monitoring of volcanic 1221 the measurement station. Both AQMesh pods
1166 gas and/or aerosol concentrations in downwind 1222 had a successful ratio of true-positives to false-
1167 communities, the residents can react to the fluc- 1223 positives, with lower percentages of identifying
1168 tuating presence of the plume. During periods 1224 the plume’s presence when meteorological data
1169 of extreme degassing from Kı̄lauea volcano in 1225 suggested the plume was elsewhere. El Crucero
1170 Hawaii in 2018, official government advice in- 1226 and Pacaya AQMesh pods were least effective at
1171 cluded remaining indoors, closing doors and win- 1227 identifying the plume’s presence via means of cor-
1172 dows and recirculating air within the building 1228 relations between SO2 and PM. No SO2 to PM cor-
1173 [Hawaii Emergency Management Agency, 2018]. 1229 relations were identified from data collected at
1174 In countries such as Nicaragua, buildings are not 1230 Pacaya station, and El Crucero’s AQMesh pod pos-
1175 commonly airtight and so mitigation strategies 1231 itively identified the presence of the plume only
1176 would focus more on avoiding excessive physi- 1232 4% of the time, and with an equal percentage of
1177 cal activity during especially high levels of ex- 1233 false-positives. The inefficiency of El Crucero and
1178 posure [Pohl 1998; Williams-Jones and Rymer 1234 Pacaya stations may be as a result of the larger dis-
1179 2015; IVHHN, 2020]. These mitigation strate- 1235 tance from the volcanic source point, providing a
1180 gies work best when there is clear communica- 1236 greater potential for bias in the determination of
1181 tion to the public regarding the concentration of 1237 plume presence from meteorological conditions.
1182 volcanic gas and aerosol which they are being ex- 1238 Analysis of the particulate data indicated that
1183 posed to in real-time. This most effectively works 1239 both near-field and far-field stations can be suit-
1184 either with a well-maintained ground-based mon- 1240 able for identifying an increase in particles dur-
1185 itoring network with readily-accessible real-time 1241 ing plume-present conditions. El Panama sta-
1186 data, or with a model forecast capable of accu- 1242 tion recorded exceedence events above the back-
1187 rately predicting the movement of the volcanic 1243 ground norm 39.5% of the measurement time
1188 plume and allowing communication to the com- 1244 for PM2.5 under plume-present conditions, as op-
1189 munity members of their level of volcanic expo- 1245 posed to 21.2% of the time for plume-absent con-
1190 sure on any given day. Although the AQMesh 1246 ditions. El Crucero similarly had a higher fre-
1191 network temporarily installed by the UNRESP 1247 quency of exceeding the background norm un-
1192 project in Nicaragua was not able to provide a 1248 der plume-present conditions (47.5% for PM2.5 )
1193 real-time warning system, it would be highly ben- 1249 as opposed to plume-absent conditions (5.5% for
1194 eficial to the communities surrounding the vol- 1250 PM2.5 ). However, both Pacaya and Rigoberto
1195 cano for such a monitoring system to be imple- 1251 measurement stations showed small variations in
1196 mented in the future. 1252 the exceedences above background norms (25.0%
1253 for plume-present PM2.5 and 20.3% for plume-
1197 6 Conclusions 1254 absent PM2.5 at Pacaya, and 14.3% for plume-
1255 present PM2.5 and 16.3% for plume-absent PM2.5
1198 A network of five AQMesh pods was installed 1256 at Rigoberto). At Rigoberto the low variability
1199 in Nicaragua by the UNRESP project between 1257 in particulates is suggested to be caused by the
1200 February and August 2017. The network data 1258 reasonably close proximity to a domestic cooking
1201 were analysed to assess the pods’ effectiveness at 1259 fire.
1202 recognising the presence of volcanic plume at the 1260 The AQMesh pods which were installed in
1203 measurement sites. Presence of volcanic plume 1261 Nicaragua were originally designed for monitor-
1204 at measurement stations was determined from 1262 ing air quality in urban and commercial environ-

Page 18
1263 ments, and were severely affected by the volcanic 1309 Masaya volcano (Nicaragua), 2014–2017. Geo-
1264 environment in this study. All AQMesh pods re- 1310 chemistry, Geophysics, Geosystems, 19(2):496–
1265 quired replacement parts to be installed due to in- 1311 515. DOI: https://agupubs.onlinelibrary.
1266 strument failure or corrosion of key components, 1312 wiley.com/doi/full/10.1002/2017GC007227.
1267 and there was a high frequency of AQMesh pods
1268 going offline. The AQMesh pods provide a rela- 1313 Allen, A., Oppenheimer, C., Ferm, M., Baxter,
1269 tively low-cost opportunity for monitoring down- 1314 P., Horrocks, L., Galle, B., McGonigle, A., and
1270 wind concentrations of volcanic gas and aerosol, 1315 Duffell, H. (2002). Primary sulfate aerosol
1271 in comparison to reference-grade instrument net- 1316 and associated emissions from Masaya Volcano,
1272 works. However an alternate monitoring network 1317 Nicaragua. Journal of Geophysical Research: At-
1273 allowing concentrations of particulates to be mea- 1318 mospheres, 107(D23):ACH–5. DOI: https://
1274 sured at a significantly lower cost could be pro- 1319 doi.org/10.1029/2002JD002120.
1275 vided by PurpleAir sensors. Countries such as 1320 AQMesh (2021a). Data Access. https://www.
1276 Nicaragua with persistently degassing volcanoes 1321 aqmesh.com/products/data-access/. (Ac-
1277 could greatly benefit from installation of some 1322 cessed 27/02/2021).
1278 form of permanent monitoring network to track
1279 volcanic plume concentrations in downwind com- 1323 AQMesh (2021b). Technical Specifications.
1280 munities in real-time. 1324 https://www.aqmesh.com/products/
1325 technical-specification/. (Accessed
1326 27/02/2021).
1281 Acknowledgements
1327 ATSDR (1998). Agency for Toxic Substances and
1282 We thank ... ? INETER? Community members, es- 1328 Disease Registry; public health statement sulfur
1283 pecially in El Panama? The people who owned the 1329 dioxide. 7446-09-5.
1284 land/buildings were the AQMesh were installed?
1285 Like the Susi Syke clinic? 1330 Austin, C. C., Roberge, B., and Goyer, N. (2006).
1331 Cross-sensitivities of electrochemical detectors
1332 used to monitor worker exposures to airborne
1286 Author contributions 1333 contaminants: False positive responses in the
1334 absence of target analytes. Journal of Environ-
1287 RCWW performed the data analysis and wrote the 1335 mental Monitoring, 8(1):161–166. DOI: https:
1288 original draft. MP, EI, TR, and AS contributed to 1336 //doi.org/10.1039/B510084D.
1289 data interpretation. All co-authors contributed to
1290 draft review and editing. 1337 Balmes, J., Fine, J., and Sheppard, D. (1987).
1338 Symptomatic bronchoconstriction after short-
1339 term inhalation of sulfur dioxide. American Re-
1291 Funding 1340 view of Respiratory Disease, 136(5). DOI: https:
1341 //doi.org/10.1164/ajrccm/136.5.1117.
1292 RCWW is funded by the Leeds-York Natural En-
1293 vironment Research Council (NERC) Doctoral 1342 Baxter, P., Stoiber, R., and Williams, S. (1982).
1294 Training Partnership (DTP) NE/L002574/1, in 1343 Volcanic gases and health: Masaya volcano,
1295 CASE partnership with the Icelandic Meteorologi- 1344 Nicaragua. The Lancet. DOI: https://doi.org/
1296 cal Office. NE/R009465/1 "Unseen but not unfelt: 1345 10.1016/S0140-6736(82)91109-6.
1297 resilience to persistent volcanic emissions (UN-
1298 RESP)".... 1346 Buizza, R., Houtekamer, P., Pellerin, G., Toth, Z.,
1347 Zhu, Y., and Wei, M. (2005). A comparison of
1348 the ECMWF, MSC, and NCEP global ensemble
1299 Data availability 1349 prediction systems. Monthly Weather Review,
1350 133(5):1076–1097. DOI: https://doi.org/10.
1300 Authors should direct readers to an open access 1351 1175/MWR2905.1.
1301 repository such as figshare or Github, where data
1302 are made available. 1352 Burton, M., Oppenheimer, C., Horrocks, L.,
1353 and Francis, P. (2000). Remote sensing of
1354 CO2 and H2 O emission rates from Masaya
1303 References 1355 volcano, Nicaragua. Geology, 28(10):915–
1356 918. DOI: https://doi.org/Doi10.1130/
1304 Aiuppa, A., de Moor, J. M., Arellano, S., Cop- 1357 0091-7613(2000)28<915:Rsocah>2.0.Co;2.
1305 pola, D., Francofonte, V., Galle, B., Giu-
1306 dice, G., Liuzzo, M., Mendoza, E., Sabal- 1358 Butwin, M. K., von Löwis, S., Pfeffer, M. A., and
1307 los, A., et al. (2018). Tracking forma- 1359 Thorsteinsson, T. (2019). The effects of vol-
1308 tion of a lava lake from ground and space: 1360 canic eruptions on the frequency of particulate

Page 19
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

1361 matter suspension events in Iceland. Journal 1414 Delmelle, P., Baxter, P., Beaulieu, A., Burton,
1362 of Aerosol Science, 128:99–113. DOI: https:// 1415 M., Francis, P., Garcia-Alvarez, J., Horrocks,
1363 doi.org/10.1016/j.jaerosci.2018.12.004. 1416 L., Navarro, M., Oppenheimer, C., Rothery, D.,
1417 et al. (1999). Origin, effects of Masaya vol-
1364 Cadle, R., Wartburg, A., and Grahek, P. (1971). 1418 cano’s continued unrest probed in Nicaragua.
1365 The proportion of sulfate to sulfur dioxide 1419 Eos, Transactions American Geophysical Union,
1366 in Kı̄lauea Volcano fume. Geochimica et Cos- 1420 80(48):575–581. DOI: https://doi.org/10.
1367 mochimica Acta, 35(5):503–507. DOI: https: 1421 1029/EO080i048p00575.
1368 //doi.org/10.1016/0016-7037(71)90046-9.
1422 Delmelle, P., Stix, J., Baxter, P., Garcia-Alvarez,
1369 Carlsen, H. K., Ilyinskaya, E., Baxter, P., Schmidt, 1423 J., and Barquero, J. (2002). Atmospheric dis-
1370 A., Thorsteinsson, T., Pfeffer, M., Barsotti, S., 1424 persion, environmental effects and potential
1371 Dominici, Francesca Finnbjornsdottir, R. G., 1425 health hazard associated with the low-altitude
1372 Jóhannsson, T., , Aspelund, T., Gislason, T., 1426 gas plume of Masaya volcano, Nicaragua. Bul-
1373 Valdimarsdóttir, U., Briem, H., and Gudna- 1427 letin of Volcanology, 64(6):423–434. DOI:
1374 son, T. (2021a). Increased respiratory morbid- 1428 10.1007/s00445-002-0221-6.
1375 ity associated with exposure to a mature vol-
1376 canic plume from a large Icelandic fissure erup- 1429 Duffell, H. J., Oppenheimer, C., Pyle, D. M., Galle,
1377 tion. Nature Communications, 12. DOI: https: 1430 B., McGonigle, A. J., and Burton, M. R. (2003).
1378 //doi.org/10.1038/s41467-021-22432-5. 1431 Changes in gas composition prior to a minor ex-
1432 plosive eruption at Masaya volcano, Nicaragua.
1379 Carlsen, H. K., Valdimarsdóttir, U., Briem, H., 1433 Journal of Volcanology and Geothermal Research,
1380 Dominici, F., Finnbjornsdottir, R. G., Jóhanns- 1434 126(3-4):327–339. DOI: https://doi.org/10.
1381 son, T., Aspelund, T., Gislason, T., and Gud- 1435 1016/S0377-0273(03)00156-2.
1382 nason, T. (2021b). Severe volcanic SO2 ex-
1383 posure and respiratory morbidity in the Ice- 1436 Environmental Protection Agency (2016).
1384 landic population–a register study. Environ- 1437 List of designated reference and equiv-
1385 mental Health, 20(1):1–12. DOI: https://doi. 1438 alent methods. National Exposure Re-
1386 org/10.1186/s12940-021-00698-y. 1439 search Laboratory. Available at: https:
1440 //www.epa.gov/sites/production/files/
1387 Cohen, A. J., Ross Anderson, H., Ostro, B., 1441 2019-08/documents/designated_reference_
1388 Pandey, K. D., Krzyzanowski, M., Künzli, N., 1442 and-equivalent_methods.pdf.
1389 Gutschmidt, K., Pope, A., Romieu, I., Samet,
1390 J. M., et al. (2005). The global burden of dis- 1443 Galle, B., Oppenheimer, C., Geyer, A., Mc-
1391 ease due to outdoor air pollution. Journal of 1444 Gonigle, A. J., Edmonds, M., and Horrocks,
1392 Toxicology and Environmental Health, Part A, 1445 L. (2003). A miniaturised ultraviolet spec-
1393 68(13-14):1301–1307. DOI: https://doi.org/ 1446 trometer for remote sensing of SO2 fluxes: a
1394 10.1080/15287390590936166. 1447 new tool for volcano surveillance. Journal
1448 of Volcanology and Geothermal Research, 119(1-
1395 CRI (2004). The Centre for Research Informa- 1449 4):241–254. DOI: https://doi.org/10.1016/
1396 tion, health effects of project shad chemical 1450 S0377-0273(02)00356-6.
1397 agent: sulfur dioxide [cas 7446-09-5]. National
1451 Global Volcanism Program (2021). Re-
1398 Academies.
1452 port on Masaya (Nicaragua) Bulletin of
1399 Crilley, L. R., Shaw, M., Pound, R., Kramer, L. J., 1453 the Global Volcanism Network. https:
1400 Price, R., Young, S., Lewis, A. C., and Pope, 1454 //volcano.si.edu/volcano.cfm?vn=344100.
1401 F. D. (2018). Evaluation of a low-cost opti- 1455 (Accessed 03/03/2021).
1402 cal particle counter (Alphasense OPC-N2) for 1456 Hawaii Emergency Management Agency (2018).
1403 ambient air monitoring. Atmospheric Measure- 1457 How to cope with hazardous volcanic gas
1404 ment Techniques, pages 709–720. DOI: https: 1458 emissions. http://dod.hawaii.gov/hiema/
1405 //doi.org/10.5194/amt-11-709-2018. 1459 how-to-cope-with-hazardous-volcanic-gas-emissions/.
1406 Crilley, L. R., Singh, A., Kramer, L. J., Shaw, M. D., 1460 (Accessed 15/4/2021).
1407 Alam, M. S., Apte, J. S., Bloss, W. J., Hilde- 1461 Holgate, S. (2017). "every breath we take: the
1408 brandt Ruiz, L., Fu, P., Fu, W., et al. (2020). Ef- 1462 lifelong impact of air pollution" - a call for ac-
1409 fect of aerosol composition on the performance 1463 tion. Clinical Medicine, 17(1):8–12. DOI: https:
1410 of low-cost optical particle counter correction 1464 //doi.org/10.7861/clinmedicine.17-1-8.
1411 factors. Atmospheric Measurement Techniques,
1412 13(3):1181–1193. DOI: https://doi.org/10. 1465 Ilyinskaya, E., Mason, E., Wieser, P., Holland,
1413 5194/amt-13-1181-2020. 1466 L., Liu, E., Mather, T., Edmonds, M., Whitty,

Page 20
1467 R., Elias, T., Nadeau, P., Schneider, D., Mc- 1521 Lim, S. S., Vos, T., Flaxman, A. D., Danaei, G.,
1468 Quaid, J., Allen, S., Harvey, J., Oppenheimer, C., 1522 Shibuya, K., Adair-Rohani, H., AlMazroa, M. A.,
1469 Kern, C., and Damby, D. (2021). Rapid metal 1523 Amann, M., Anderson, H. R., Andrews, K. G.,
1470 pollutant deposition from the volcanic plume 1524 et al. (2012). A comparative risk assessment
1471 of Kı̄lauea, Hawai‘i,. Communications Earth 1525 of burden of disease and injury attributable to
1472 and Environment. DOI: https://doi.org/10. 1526 67 risk factors and risk factor clusters in 21
1473 1038/s43247-021-00146-2. 1527 regions, 1990–2010: a systematic analysis for
1528 the Global Burden of Disease Study 2010. The
1474 Ilyinskaya, E., Schmidt, A., Mather, T. A., Pope,
1529 lancet, 380(9859):2224–2260. DOI: https://
1475 F. D., Witham, C., Baxter, P., Jóhannsson, T., Pf-
1530 doi.org/10.1016/S0140-6736(12)61766-8.
1476 effer, M., Barsotti, S., Singh, A., et al. (2017).
1477 Understanding the environmental impacts of 1531 Longo, B., Rossignol, A., and Green, J. (2008).
1478 large fissure eruptions: Aerosol and gas emis- 1532 Cardiorespiratory health effects associated with
1479 sions from the 2014–2015 Holuhraun erup- 1533 sulphurous volcanic air pollution. Public
1480 tion (Iceland). Earth and Planetary Science Let- 1534 Health, 122(8):809–20. DOI: https://doi.
1481 ters, 472:309–322. DOI: https://doi.org/10. 1535 org/10.1016/j.puhe.2007.09.017.
1482 1016/j.epsl.2017.05.025.
1536 Longo, B. M. (2013). Adverse health effects associ-
1483 International Volcanic Health Hazard Network
1537 ated with increased activity at Kı̄lauea Volcano:
1484 (2020). The health hazards of volcanic and
1538 A repeated population-based survey. ISRN Pub-
1485 geothermal gases: A guide for the public.
1539 lic Health, 2013. DOI: https://doi.org/10.
1486 https://www.ivhhn.org/information/
1540 1155/2013/475962.
1487 health-impacts-volcanic-gases#
1488 Howtolivewithvolcanicemissions. (Ac- 1541 Loughlin, S., Aspinall, W., Vye-Brown, C.,
1489 cessed 12/4/2021). 1542 Baxter, P., Braban, C., Hort, M., Schmidt,
1490 Jayaratne, R., Liu, X., Thai, P., Dunbabin, M., 1543 A., Thordarson, T., and Witham, C. (2012).
1491 and Morawska, L. (2018). The influence of hu- 1544 Large-magnitude fissure eruptions in Iceland:
1492 midity on the performance of a low-cost air 1545 source characterisation. BGS Open File Report,
1493 particle mass sensor and the effect of atmo- 1546 OR/12/098.
1494 spheric fog. Atmospheric Measurement Tech-
1547 Mason, E., Wieser, P., Liu, E., Edmonds, M.,
1495 niques, 11(8):4883–4890. DOI: https://doi.
1548 Ilyinskaya, E., Whitty, R., Mather, T., Elias, T.,
1496 org/10.5194/amt-11-4883-2018.
1549 Nadeau, P., Wilkes, T., et al. (2021). Volatile
1497 Kelly, K., Whitaker, J., Petty, A., Widmer, C., Dyb- 1550 metal emissions from volcanic degassing and
1498 wad, A., Sleeth, D., Martin, R., and Butterfield, 1551 lava-seawater interactions at Kı̄lauea Volcano,
1499 A. (2017). Ambient and laboratory evaluation 1552 Hawai’i. Communications Earth and Environ-
1500 of a low-cost particulate matter sensor. Envi- 1553 ment. DOI: 10.1038/s43247-021-00146-2.
1501 ronmental Pollution, 221:491–500. DOI: https:
1502 //doi.org/10.1016/j.envpol.2016.12.039. 1554 Mather, T., Allen, A., Oppenheimer, C., Pyle,
1555 D., and McGonigle, A. (2003). Size-resolved
1503 Krogh, A. (2021). Snazzy maps. https:// 1556 characterisation of soluble ions in the particles
1504 snazzymaps.com/. (Accessed 20/01/2021). 1557 in the tropospheric plume of Masaya volcano,
1558 Nicaragua: origins and plume processing. Jour-
1505 Lambert, G., Le Cloarec, M., and Pennisi, M.
1559 nal of Atmospheric Chemistry, 46(3):207–237.
1506 (1988). Volcanic output of SO2 and trace met-
1560 DOI: 10.1023/A:1026327502060.
1507 als: a new approach. Geochimica et Cosmochim-
1508 ica Acta, 52(1):39–42. DOI: https://doi.org/ 1561 Mather, T. A. (2015). Volcanoes and the environ-
1509 10.1016/0016-7037(88)90054-3. 1562 ment: Lessons for understanding Earth’s past
1510 Langmann, B. (2014). On the role of climate forc- 1563 and future from studies of present-day volcanic
1511 ing by volcanic sulphate and volcanic ash. Ad- 1564 emissions. Journal of Volcanology and Geother-
1512 vances in Meteorology, 2014:1–17. DOI: https: 1565 mal Research, 304:160–179. DOI: https://doi.
1513 //doi.org/10.1155/2014/340123. 1566 org/10.1016/j.jvolgeores.2015.08.016.

1514 Lewis, A. C., Lee, J. D., Edwards, P. M., Shaw, 1567 Mather, T. A., McCabe, J. R., Rai, V. K., Thiemens,
1515 M. D., Evans, M. J., Moller, S. J., Smith, 1568 M. H., Pyle, D. M., Heaton, T. H. E., Sloane, H. J.,
1516 K. R., Buckley, J. W., Ellis, M., Gillot, S. R., 1569 and Fern, G. R. (2006a). Oxygen and sulfur iso-
1517 et al. (2016). Evaluating the performance of 1570 topic composition of volcanic sulfate aerosol at
1518 low cost chemical sensors for air pollution re- 1571 the point of emission. Journal of Geophysical Re-
1519 search. Faraday discussions, 189:85–103. DOI: 1572 search, 111. DOI: https://doi.org/10.1029/
1520 10.1039/C5FD00201J. 1573 2005JD006584.

Page 21
Masaya volcano, Nicaragua: How effective are low-cost air quality monitors for identifying SO2 and PM
volcanic signatures in downwind locations? Whitty et al., 2021

1574 Mather, T. A., Pyle, D., Tsanev, V., McGo- 1627 and deposition of Indonesian volcanic
1575 nigle, A., Oppenheimer, C., and Allen, A. 1628 emissions. Atmospheric Chemistry and
1576 (2006b). A reassessment of current volcanic 1629 Physics, 6(9):2525–2537. DOI: https:
1577 emissions from the Central American arc with 1630 //doi.org/10.5194/acp-6-2525-2006.
1578 specific examples from Nicaragua. Journal
1631 Pfeffer, M., Rietmeijer, F., Brearley, A., and Fis-
1579 of volcanology and geothermal Research, 149(3-
1632 cher, T. P. (2006b). Electron microbeam anal-
1580 4):297–311. DOI: https://doi.org/10.1016/
1633 yses of aerosol particles from the plume of Poás
1581 j.jvolgeores.2005.07.021.
1634 Volcano, Costa Rica and comparison with equi-
1582 McBirney, A. R. (1956). The Nicaraguan vol- 1635 librium plume chemistry modeling. Journal
1583 cano Masaya and its caldera. Eos, Transactions 1636 of volcanology and geothermal research, 152(1-
1584 American Geophysical Union, 37(1):83–96. DOI: 1637 2):174–188. DOI: https://doi.org/10.1016/
1585 https://doi.org/10.1029/TR037i001p00083. 1638 j.jvolgeores.2005.10.009.

1586 Mead, M., Popoola, O., Stewart, G., Landshoff, 1639 Pohl, H. R. (1998). Toxicological profile for sul-
1587 P., Calleja, M., Hayes, M., Baldovi, J., McLeod, 1640 fur dioxide. U.S. Department for Health and
1588 M., Hodgson, T., Dicks, J., et al. (2013). The 1641 Human Services. Available at: https://www.
1589 use of electrochemical sensors for monitoring 1642 atsdr.cdc.gov/toxprofiles/tp116.pdf.
1590 urban air quality in low-cost, high-density net-
1643 PurpleAir (2019). PurpleAir: Air Quality Mon-
1591 works. Atmospheric Environment, 70:186–203.
1644 itoring. https://www.purpleair.com/. (Ac-
1592 DOI: https://doi.org/10.1016/j.atmosenv.
1645 cessed 26/10/2018).
1593 2012.11.060.
1646 Read, K. (2018). Report of Calibration
1594 Miller, V. (2004). Health Effects of Project SHAD 1647 Work. https://github.com/ncasuk/
1595 Chemical Agent: Sulfur Dioxide, Prepared for 1648 ncas-cozi-lab-additionaldata/blob/main/
1596 the National Academies by The Center for Re- 1649 20180117_Certificate_SO2_43I_TRACE.pdf.
1597 search Information. Contract No. IOM-2794-04-
1598 001, pages 1–58. 1650 Rotstayn, L. D. and Lohmann, U. (2002). Simula-
1651 tion of the tropospheric sulfur cycle in a global
1599 Molteni, F., Buizza, R., Palmer, T. N., and 1652 model with a physically based cloud scheme.
1600 Petroliagis, T. (1996). The ECMWF ensem- 1653 Journal of Geophysical Research: Atmospheres,
1601 ble prediction system: Methodology and val- 1654 107(D21). DOI: https://doi.org/10.1029/
1602 idation. Quarterly journal of the royal me- 1655 2002JD002128.
1603 teorological society, 122(529):73–119. DOI:
1604 10.21957/jl1b71oqj. 1656 Rymer, H., de Vries, B. v. W., Stix, J., and Williams-
1657 Jones, G. (1998). Pit crater structure and pro-
1605 Nadeau, P. A. and Williams-Jones, G. (2009). 1658 cesses governing persistent activity at Masaya
1606 Apparent downwind depletion of volcanic 1659 Volcano, Nicaragua. Bulletin of Volcanology,
1607 SO2 flux — lessons from Masaya Volcano, 1660 59(5):345–355. DOI: 10.1007/s004450050196.
1608 Nicaragua. Bulletin of Volcanology, 71(4):389–
1609 400. DOI: DOI 10.1007/s00445-008-0251-9. 1661 Saxena, P. and Seigneur, C. (1987). On the oxida-
1662 tion of SO2 to sulfate in atmospheric aerosols.
1610 Oppenheimer, C., Francis, P., and Stix, J. (1998). 1663 Atmospheric Environment (1967), 21(4):807–
1611 Depletion rates of sulfur dioxide in tropo- 1664 812.
1612 spheric volcanic plumes. Geophysical Research
1613 Letters, 25(14):2671–2674. DOI: https://doi. 1665 Sayahi, T., Butterfield, A., and Kelly, K. (2019).
1614 org/10.1029/98GL01988. 1666 Long-term field evaluation of the Plantower
1667 PMS low-cost particulate matter sensors. Envi-
1615 Oppenheimer, C. and McGonigle, J. (2004). Ex- 1668 ronmental Pollution, 245:932–940. DOI: https:
1616 ploiting ground-based optical sensing tech- 1669 //doi.org/10.1016/j.envpol.2018.11.065.
1617 nologies for volcanic gas surveillance. Annals
1618 Geophysics, 47(4). DOI: https://doi.org/10. 1670 Schlesinger, R., Kunzli, N., Hidy, G., Gotschi, T.,
1619 4401/ag-3353. 1671 and Jerrett, M. (2006). The health relevance of
1672 ambient particulate matter characteristics: co-
1620 Pattantyus, A. K., Businger, S., and Howell, S. G. 1673 herence of toxicological and epidemiological in-
1621 (2018). Review of sulfur dioxide to sulfate 1674 ferences. Inhalation toxicology, 18(2):95–125.
1622 aerosol chemistry at Kı̄lauea Volcano, Hawai‘i. 1675 DOI: 10.1080/08958370500306016.
1623 Atmospheric Environment, 185:262–271. DOI:
1624 10.1016/j.atmosenv.2018.04.055. 1676 Schlesinger, R. B. (1985). Effects of inhaled acids
1677 on respiratory tract defense mechanisms. En-
1625 Pfeffer, M., Langmann, B., and Graf, 1678 vironmental health perspectives, 63:25–38. DOI:
1626 H.-F. (2006a). Atmospheric transport 1679 https://doi.org/10.1289/ehp.856325.

Page 22
1680 Schmidt, A., Leadbetter, S., Theys, N., Carboni, 1734 at: https://www.thermofisher.com/order/
1681 E., Witham, C., Stevenson, J., Birch, C., Thor- 1735 catalog/product/43I#/43I.
1682 darson, T., Turnock, S., Barsotti, S., Delaney,
1683 L., Feng, W., Grainger, R., Hort, M., Höskulds- 1736 van Manen, S. M. (2014). Perception of a chronic
1684 son, A., Ialongo, I., Ilyinskaya, E., Jóhanns- 1737 volcanic hazard: persistent degassing at Masaya
1685 son, T., Kenny, P., Mather, T., Richards, N., 1738 volcano, Nicaragua. Journal of Applied Volcanol-
1686 and Shepherd, J. (2015). Satellite detection, 1739 ogy, 3(1):9. DOI: 10.1186/s13617-014-0009-3.
1687 long-range transport, and air quality impacts 1740 von Glasow, R., Bobrowski, N., and Kern, C.
1688 of volcanic sulfur dioxide from the 2014–2015 1741 (2009). The effects of volcanic eruptions on at-
1689 flood lava eruption at Bárðarbunga (Iceland). 1742 mospheric chemistry. Chemical Geology, 263(1-
1690 Journal of Geophysical Research: Atmospheres, 1743 4):131–142. DOI: https://doi.org/10.1016/
1691 120(18):9739–9757. DOI: https://doi.org/ 1744 j.chemgeo.2008.08.020.
1692 10.1002/2015JD023638.
1745 Whitty, R. C., Ilyinskaya, E., Mason, E., Wieser,
1693 Sousan, S., Koehler, K., Hallett, L., and Pe- 1746 P. E., Liu, E. J., Schmidt, A., Roberts, T., Pfeffer,
1694 ters, T. M. (2016a). Evaluation of the Al- 1747 M. A., Brooks, B., Mather, T. A., et al. (2020).
1695 phasense optical particle counter (OPC-N2) 1748 Spatial and Temporal Variations in SO2 and
1696 and the Grimm portable aerosol spectrometer 1749 PM2.5 Levels Around Kı̄lauea Volcano, Hawai’i
1697 (PAS-1.108). Aerosol Science and Technology, 1750 During 2007–2018. Frontiers in Earth Science,
1698 50(12):1352–1365. DOI: https://doi.org/10. 1751 8:36. DOI: https://doi.org/10.3389/feart.
1699 1080/02786826.2016.1232859. 1752 2020.00036.
1700 Sousan, S., Koehler, K., Thomas, G., Park, J. H., 1753 Williams-Jones, G. and Rymer, H. (2015). Hazards
1701 Hillman, M., Halterman, A., and Peters, T. M. 1754 of volcanic gases. The Encyclopedia of Volcanoes,
1702 (2016b). Inter-comparison of low-cost sensors 1755 2:985–992. DOI: https://doi.org/10.1016/
1703 for measuring the mass concentration of occu- 1756 B978-0-12-385938-9.00057-2.
1704 pational aerosols. Aerosol Science and Technol-
1705 ogy, 50(5):462–473. DOI: https://doi.org/ 1757 Williams-Jones, G., Rymer, H., and Rothery,
1706 10.1080/02786826.2016.1162901. 1758 D. A. (2003). Gravity changes and passive
1759 SO2 degassing at the Masaya caldera complex,
1707 Stockwell, W. R. and Calvert, J. G. (1983). 1760 Nicaragua. Journal of Volcanology and Geother-
1708 The mechanism of the HO-SO2 reaction. 1761 mal Research, 123(1-2):137–160. DOI: https://
1709 Atmospheric Environment (1967), 17(11):2231– 1762 doi.org/10.1016/S0377-0273(03)00033-7.
1710 2235. DOI: https://doi.org/10.1016/
1711 0004-6981(83)90220-2.
1712 Stoiber, R. E., Williams, S. N., and Huebert, B. J.
1713 (1986). Sulfur and halogen gases at Masaya
1714 caldera complex, Nicaragua: total flux and vari-
1715 ations with time. Journal of Geophysical Re-
1716 search: Solid Earth, 91(B12):12215–12231. DOI:
1717 https://doi.org/10.1029/JB091iB12p12215.
1718 Tam, E., Miike, R., Labrenz, S., Sutton, A. J., Elias,
1719 T., Davis, J., Chen, Y.-L., Tantisira, K., Dockery,
1720 D., and Avol, E. (2016). Volcanic air pollution
1721 over the Island of Hawai‘i: Emissions, disper-
1722 sal, and composition. association with respira-
1723 tory symptoms and lung function in Hawai‘i Is-
1724 land school children. Environment international,
1725 92:543–552. DOI: https://doi.org/10.1016/
1726 j.envint.2016.03.025.
1727 Tamburello, G. (2015). Ratiocalc: Software for
1728 processing data from multicomponent volcanic
1729 gas analyzers. Computers & Geosciences, 82:63–
1730 67. DOI: https://doi.org/10.1016/j.cageo.
1731 2015.05.004.
1732 Thermo Scientific (2010). Model 43i sulphur diox-
1733 ide analyser - product specifications. . Available

Page 23

You might also like