Remote sensing and identification of volcanic plumes using fixed‐wing UAVs over Volcán de Fuego, Guatemala

This paper describes a series of proof‐of‐concept Beyond Visual Line Of Sight unmanned aerial vehicle flights which reached a range of up to 9 km and an altitude of 4,410 m Above Mean Sea Level over Volcán de Fuego in Guatemala, interacting with the volcanic plume on multiple occasions across a range of different conditions. Volcán de Fuego is an active volcano which emits gas and ash regularly, causing disruption to airlines operating from the international airport 50 km away and impacting the lives of the local population. Collection of data from within the plume develops scientists’ understanding of the composition of the volcano's output and is of use to scientists, aviation, and hazard management groups alike. This paper presents preliminary results of multiple plume interceptions with multiple aircraft, carrying a variety of sensors. A plume‐detection metric is introduced, which uses a combination of flight data and atmospheric sensor data to identify flight through a volcanic plume. Future work will develop the automation of plume tracking such that reliable scientific data sets can be gathered in a robust manner.

accuracy of satellite retrievals and reliability of dispersion model predictions (Western, Watson, & Francis, 2015). Representative ash samples are therefore critical for the effective use of both airspace and industry management tools, to respond to volcanic events accordingly. Additionally, the knowledge gained by analysing such samples can also help mitigate the effects of large eruptions on the local population by better informing local decision makers.

| Volcán de Fuego
Volcán de Fuego (henceforth referred to as Fuego) is an active volcano in Guatemala with a summit altitude of approximately 3,800 m Above Mean Sea Level (AMSL). Short-term small explosions occur multiple times every hour, with larger, more sustained, eruptions every 3-5 weeks. Occasional major eruptions (e.g., June 2018, September 2012, and October 1974 cause large-scale changes to the area's topography and have a significant impact on the local population. Over 1M people live within a 30 km radius of the volcano (Smithsonian Institute, 2002), and the country's international airport is <50 km away, so the volcano poses a large risk to the Guatemalan population, aviation industry, and economy. Whereas other active volcano craters are relatively accessible for scientists, the large volcanic projectiles that are emitted on ballistic trajectories make Fuego a particularly hard volcano from which to collect samples.
Ground-based collection of ash that has fallen out of the plume is common and straightforward, however these samples cannot have a PSD representative of the in-plume PSD. Airborne ash collection from within the plume would have a more representative PSD and poses an interesting engineering problem because novel methods must be used. shows the visible part of the plume from two separate eruptions. In conditions, such as these, it is easy to confirm flight through the plume using the visual cameras on-board the unmanned aerial vehicles (UAVs). The weather around Fuego can be unpredictable, and cloud often arrives between the ground and summit altitude around the middle of the day. Even in these conditions the plume is still visible on camera as it shows up as darker than the normal clouds.

| UAVs for volcanic and atmospheric research
UAVs enable a wide range of operations that would otherwise not be possible due to restrictions on human safety, physical limitations, and do so at a lower cost than most alternative means of gathering data. Everaerts (2008), Wegener (2004), and Klemas (2015) present early papers on the potential use of UAVs for scientific missions, with Everaerts highlighting the usefulness of UAVs for scientific remote sensing and mapping due to the low cost and ease of access to platforms. Klemas compares UAVs for remote sensing over coastal areas with the manned aircraft that were previously used at great expense. UAVs give excellent access to extreme environments, as demonstrated by Di Stefano who used a multicopter to monitor and gather data over the Lusi mud crater in Indonesia (Di Stefano et al., 2017). Ramanathan et al. (2007) pioneered the use of lightweight fixedwing UAVs to investigate atmospheric phenomena, equipping three large fixed-wing aircraft with aerosol, soot, and solar radiation instrumentation to measure the heating effect of brown carbon layers between 0.5 and 3 km above the Indian Ocean. They were flown in stacked formation for simultaneous data collection from different altitudes. The additional development of a turbulent flux measurement system for UAVs by R. M. Thomas et al. (2012) enabled further investigations above the Indian Ocean, including the discovery that solar absorption by black carbon particles suppresses boundary layer turbulence (Wilcox et al., 2016). As Thomas notes in the 2012 paper, instruments for taking representative measurements of atmospheric phenomena often need to be out of the boundary layer of the aircraft. These instruments can be a significant component of the overall takeoff mass, leading to challenges with balance and flight control when mounted as required. Villa, Gonzalez, Miljievic, Ristovski, and Morawska (2016)  | 1193 desirable method of collecting air data, their paper highlights a few key challenges, including a necessity for sensors that are both small and suitably accurate, and overcoming policies/regulations, such as import systems and airspace rules. These factors translate into most other applications of UAVs. Schuyler and Guzman (2017) review the various UAV options available for sampling tropospheric gases, and conclude that fixed-wing aircraft with a wingspan of under 3 m and a payload of <5 kg provide the best compromise between cost and convenience of sensor deployment. Of note here is the value associated with the hours of preparation and testing in research and development of an unmanned system, a factor which is often ignored in the planned use of UAVs for ground-breaking research. Greatwood et al. (2016Greatwood et al. ( , 2017 used multirotor UAVs to collect high-altitude nonvolcanic scientific data, sampling the atmosphere above Ascension Island at approximately 3,000 m AMSL. Large (25 kg) fixed-wing UAVs have been used by Altstädter et al. (2015) to observe ultrafine particle distributions within the atmospheric boundary layer, however these flights only flew to a maximum altitude of 1,080 m AMSL.
UAVs offer outstanding new sampling opportunities for volcanic emissions (Ogiso et al., 2016). The Handbook of UAVs discusses the application of UAVs to volcanic monitoring and sampling, giving some examples of early projects and noting that their results are still preliminary due to the harshness of the environment (Longo, Melita, Muscato, & Giudice, 2014). Gas and ash sensors have been miniaturised enough to be flown on UAVs and are able to provide real-time information and, in principle, capture and retrieve samples.
UAVs allow direct and remote measurements much closer to volcanic vents than previously possible at volcanoes, such as Fuego, leading to better characterisation of the plume. Autopilot hardware and navigation algorithms improve repeatability, which should serve to better validate satellite and ground-based remote observations. Depending on the sensor and aim of collection, different UAVs can be used; for example, sampling a single location near a vent would be suited best to a multirotor UAV, but longer flight times and higher distances can be achieved by fixed-wing UAVs so they are better suited to sampling at varying distal ranges from the crater.
The first reported use of UAVs over volcanoes for scientific purposes was by McGonigle et al. (2008) in 2007, who used a helicopter UAV with a payload capacity of 3 kg at La Fossa crater in Italy. Amici et al. (2013) report the development of multirotor and fixed-wing UAV flights over Stromboli volcano in Italy, however they focus on the successes of a hexacopter with a thermal camera onboard. Jordan (2015) presents a short summary of UAVs in geology, focusing on small multirotors, such as DJI Phantoms (DJI, Shenzhen, China) and the challenges surrounding their use in scientific fieldwork. Whereas Fuego's activity involves large ballistics, active volcanoes, such as Turrialba and Masaya in Central America, are safer to be close to, hence TakeOff/Land Points (TOLPs) can be found relatively close to the craters (Stix et al., 2018). As Stix et al. show, with minimal altitude gain and short ranges required, multirotors are ideal platforms for sensor placement in static locations for gas measurements. These works consider volcanoes which are relatively accessible and note flight times between 12 and 15 min. Some areas of interest, such as Fuego, are dangerous to approach, meaning that flight times must be longer and cruise speeds higher to reach them from a safe distance. Fixed-wing UAVs are a natural solution to this issue, as demonstrated by Pieri et al. (2013), who flew fixed-wing UAVs over Turrialba, while also flying 'tethersonde' meteorological balloons for data verification. Primarily sensing gases, these experiments validated the use of UAVs in scientific data collection over volcanoes, specifically delta-wing style fixed-wing aircraft.

| UAVs at Volcán de Fuego
With an estimated summit altitude of 3,760 m AMSL, a UAV that flies over Fuego must be equipped with the appropriate components to allow for safe, reliable, and useful operations. Paredes, Saito, Abarca, and Cuellar (2017) investigate the effects of high altitude on UAV performance, validating theories generated from background aeronautical knowledge with fixed-wing flights at over 5,500 m AMSL.
Parades concludes that energy requirements increase with target altitude, which in turn decreases flight time given a fixed capacity of energy on-board.
INSIVUMEH is the Guatemalan Institute of Seismology, Volcanology, Meteorology, and Hydrology, and work to monitor sites, such as Fuego, to better understand the underlying activity and recommend evacuations when and where necessary. Before these campaigns, the summit of Fuego had last been closely observed in 2012 by manned aircraft, according to local INSIVUMEH observers.
Although macroscale behaviour of the volcano can already be monitored with seismometers, cameras, spectrometers, and satellites, close-range behaviour and topography of the summit changes regularly and remains largely unknown, as does the make-up of the volcanic emissions. As the literature shows, UAVs provide unprecedented access to hazardous, often inaccessible, zones around volcanoes. To increase the accuracy of aviation management tools, which would allow for safer operations out of the nearby international airport, a better understanding of the PSD within the plume is desirable.
Although multirotors have their place in remote sensing, particularly where static data acquisition is advantageous, operating them over Fuego's summit involves several hours travel each day just to get to the TOLP, on a road that is impassable for much of the year. The greater range of fixed-wing vehicles relative to multirotors often allows the base of operations to be located at more favourable locations with access to shelter and power. Minimising operators' travel time and logistical issues should enable UAV-based measurements to be more effectively integrated into regular operational monitoring capabilities. The availability of various TOLPs in the area means that, given a sufficient weather window, year-round flights should be possible even if the primary TOLP is inaccessible.
Challenges associated with operating in this region include the large distances and altitude gains required to reach above summit altitude, and overcoming the low air density that is inherent to such 'hot and high' locations. The work presented here made use of a single TOLP for fixed-wing aircraft; the INSIVUMEH Observatory located just north of Panimache, approximately 8 km South-West of Fuego's summit ( Figure 2). This is where the Ground Control Station (GCS) was located for the UAV missions described in this paper.

| Automated plume detection
Fixed-wing UAVs have not been used to collect volcanic data from Fuego before. A typical mission aim is to operate for as long as possible in the section of interest in the plume, for example, to collect ash from the plume between 0.3 and 3 km from the summit, from crater altitude to approximately 500 m above.
Typical UAV flight plans require the mission path to be defined by a number of waypoints before takeoff. Communications difficulties can arise when attempting to change the mission midflight, especially at long range. Should a partial mission be received by the aircraft, unexpected behaviour may follow, in the worst case leading to loss of aircraft. This mission planning method means that, for the best chance of intercepting the plume, the plume bearing and altitude must be determined preflight. This is a challenge due to the parallax error introduced by single-point ground measurements and is subject to good weather conditions. A system of observers and instruments situated at various points around the volcano could be used, however the challenging logistics, human error, and weather conditions make this unattractive. Provided there is still sufficient signal strength for telemetry, control, and video, it is possible for the pilot to take control of the aircraft in a Fly-By-Wire (FBW) mode to navigate towards the plume visually. Although likely to succeed on a case-bycase basis, from an analysis perspective this is an uncontrolled method and inconsistent between flights. It would be preferable to have an automated plume-detection system to reliably intercept the plume. The aircraft should autonomously takeoff, find the plume, collect the appropriate data, and then return to the GCS to land. The removal of human input from the system poses an interesting problem and one that, if solved, would reduce the risk of mission failure, increase repeatability, and allow nonexperts to make use of the system. Automated takeoffs and landings have been presented on numerous occasions and are regularly used by UAV operators. The detection of plumes has been investigated by Montes, Letheren, Villa, and Gonzalez (2014) and Letheren and Montes (2016), for finding the source of forest fire plumes using multirotor aircraft. Finding the plume involves the aircraft processing real-time data to establish whether contact with the plume has been made, preferably using sensors that do not conflict with scientific instruments on-board so that independence may be maintained. Letheren's system detected the (fire) plume by sensing CO 2 , and the algorithm they implemented targeted finding the source of the plume. At Fuego, the aim would be to sample specific points in the plume rather than to find its source. The algorithm(s) used on a fixedwing aircraft for plume detection and tracking must differ from those used on multirotors because a significant forward speed must always be maintained with fixed-wing aircraft.

| Research aims
The purpose of this paper is to establish a preliminary data set for the development of a system that uses UAVs to autonomously sense and quantify aspects of volcanic plumes, and to develop a metric that could be used for reliable plume detection. Clear identification of plumes would allow tracking methods, such as those used by Letheren and Montes (2016), to be implemented, enabling reliable and consistent interaction with the plume.
The hypothesis is that it is possible to detect UAV/Volcanic Plume interaction over Volcán de Fuego using a combination of temperature, ambient relative humidity, and vertical acceleration. A combination of | 1195 these data would make up a suitable metric for determining whether or not a UAV is in a plume. Specific behaviours of the UAV can then be implemented to maintain contact with the plume and collect scientific data in a controlled and repeatable manner.
The development of a system with minimal mass and power requirements maximises the usable payload of the aircraft for scientific sensors. Temperature and humidity sensors are typically small, and the on-board autopilot has integrated accelerometers.
Additional sensing methods are available, however if it is preferable to sense and identify the plume without the excess weight if possible.
To test the hypothesis, UAVs must be flown through the plume of Fuego multiple times with a variety of sensing methods on-board, so that plume-interception data can be collected by the appropriate sensors and verified by secondary methods. Should the hypothesis be found true, the automation of plume detection is a natural next step.

| Operating environment
The INSIVUMEH Observatory is at 1,137 m AMSL which, combined with the tropical climate, leads to challenging 'hot and high' conditions. Pressure decreases with altitude and air temperature, which in turn reduces the air density. For a given amount of lift a reduction in density must be compensated for by another term, the most effective being velocity due to the exponent. Table 1 Table 2, including the avionics. Both aircraft were available as hobbyist kits and were assembled in a bespoke configuration for this project. The lack of significant vertical fins mean flying-wings must typically cruise at a relatively high airspeed for a given aircraft size, to avoid dutch roll instabilities and tip stalls. The Skywalker was chosen because the airframe offers a large payload bay and their high cruise speed suited the planned mission distances.
The Zephyr II and Skywalkers were fitted with autopilot systems running ArduPlane, an open-source code base which has been used with good results for a number of years. Despite newer versions being released, for these campaigns the version was kept consistent at 3.7.1. The avionics fitted are listed in Table 2 and were chosen for automated long-range flight. A thin iron-on coating was also applied to the aircraft frame to help ensure smooth aerodynamic surfaces, but had an additional effect of increasing the airframes' robustness.
In each aircraft two cameras were mounted in the nose with a

| Sensing methods
The methods listed below are for sensing and identifying the plume.
The merits of each will be discussed individually as a method for reliably detecting when the aircraft is within a volcanic plume.

| Turbulence
The plume is expected to be turbulent relative to ambient air due to the interaction of recently expelled hot ash and gases with the ambient atmosphere. Both the PixFalcon and PixHawk AutoPilot units are fitted with a variety of on-board sensors, which are used and logged by the flight control system. Gyroscopes and accelerometers measure rotation and acceleration around and along each aircraft axis, respectively.
Turbulence was sensed by these accelerometers, and to some extent the gyroscopes, however these sensors cannot distinguish plume turbulence from other sources of turbulence, such as clouds.
Properly quantification of turbulence requires specialist devices, such as a five-hole gust probe linked to an accurate, fast-response inertial measurement unit (IMU). These data can then be used to  Stull (2005). The nature of the operating environment meant these delicate devices would need sophisticated protection measures, which was not conducive to minimising the equipment required to sense the plume. Fortunately, vertical (Z-axis) acceleration is sensed by the IMU in the flight computer and is sufficiently representative of turbulence for the plume identification attempted here. These data could easily be incorporated into an on-board detection system as no additional mass or significant computing power would be involved.

| Temperature and relative humidity
As the plume reaches maximum altitude soon after eruption, the plume as a large entity becomes neutrally buoyant. This means that as an entire body the temperature within it is no longer significantly higher than that of the surrounding air. If the plume is flown through after this stage it is likely that any temperature change measured would be minimal. Volcanic emissions from Fuego originate from the subduction zone off the coast of Guatemala in the Pacific Ocean, which could lead to a humid plume upon expulsion into clear air. It is likely that this will vary with distal range (from the summit). If there is a significant change in either temperature or relative humidity upon entering the plume then this could be included into the plumedetection system with relative ease, because temperature and relative humidity (Temp/RH) sensors are often small and have low power requirements.
Two types of Temp/RH sensor were used, both of which are described below. One is available to buy commercially and the second was developed by a member of the team.

iMet-XQ, manufactured by International Met Systems
The sensor contains a GPS unit, bead thermistor, capacitative RH sensor, and piezoresistive pressure sensor, logging at 60 Hz. The manufacturer's specifications are given in Table 3. This was mounted inside the nose of the aircraft, with the sensor tips in the airflow entering the aircraft, or externally on the forward section of the fuselage.

Avian Meteorological Package
The Avian Meteorological Package (AMP) is a modified version of the Eagle Sensor Package (ESP) as described by R. M. Thomas et al. (2018) and Greatwood et al. (2017). The ESP evolved to use an Atmel M0 (Atmel, San Jose, CA) cortex chip on a commercially available microcontroller (Adafruit Feather), and a daughter board was constructed containing a GPS chip, accelerometers, magnetometers, BMP280 pressure sensor, the fast tip, and I 2 C connections for the P14 rapid RH sensor, which were all logged at 5 Hz.
The AMP used here is an alternative to the iMet, with greater flexibility for additional sensors and integration with UAV systems.
To ensure the sensors were in the best airflow possible, without creating disproportionate amounts of drag, it was mounted on the upper surface of the fuselage near the leading edge. The specifications are given in Table 4. Although the AMP is pending validation for sensing of absolute values, the data collected qualitatively show its potential use.

| Visual cameras
Visual identification of the plume is possible using standard RGB cameras. Set to record at a resolution of 1,920 × 1,080 pixels, the GoPro They must be analysed postflight in a laboratory environment so they are not possible to incorporate with a real-time detection system.
A customised ash-collection unit was created for mounting the SEM stubs on the vehicle in a position exposed to the airflow.
A servo-operated cover allowed the stubs to be isolated during takeoff, climb/descent, and landing to prevent contamination, and was controlled manually by the operator from the GCS. Mounted on the floor at the rear of the payload bay in the Skywalkers, this unit sat in the airflow proud of the main hatch.

| IN-PLUME-DETECTION METRIC
The following metric was developed for real-time plume sensing: t is the flight time in seconds, alt is the altitude in metres AMSL, absH is the absolute humidity, relH is the relative humidity, and z is the Z-axis vertical acceleration in m/s 2 . P inPlume in Equation (1) represents the probability of the aircraft being in a plume.
where C represents the components of the formula, the individual definitions of which are defined in Equations (3)  This initial plume-detection metric was developed during the postprocessing of the flight results, and makes the following assumptions: • The plume maintains near-constant altitude in the flight area around Fuego, such that a lower altitude limit of 3,860 m AMSL can be imposed.
• The absolute humidity of the plume is below 4.95 g/m 3 , which it was on all the processed occasions.
• The standard deviation of the accelerometer data over the past 10 s (500 points at 10 Hz) is a reasonable indicator of turbulence levels.
• The difference between the current relative humidity and the minimum value of relative humidity in the last 2 min (7,200 points at 60 Hz) is a reasonable indicator of local data peaks.

| FLIGHT RESULTS
A total of 19 flights reached altitudes over 1,900 m AMSL across the three trips, with 11 reaching plume altitude or above. Meteorological conditions were best in the morning with rapid deterioration limiting useful operating time and often requiring early return of the aircraft.
Four key flights were chosen for analysis here, each having intersected the plume independently. Details of these flights are given in Table 5.
Supporting Information Material has been submitted alongside

| Flight B
The data presented here are from the first of two important flights in November 2017 and indicate intersection with the plume using the autopilot sensors, GoPro camera, Temp/RH mounted as shown in Figure 5, and ash collection. This is significant for the development of automated plume detection using UAVs.
The bearing and altitude of the plume were estimated using ground measurements, then the flight plan designed so that a large cross-wind area was covered to maximise the chance of plume

| PLUME DETECTION AND FLIGHT ANALYSIS
In this paper the sensing and identification of volcanic plumes using fixedwing UAVs is considered. A volcanic plume is the mixture of gases and ash emitted by an eruption, however for sensing purposes the boundary is hard to define, with gases diffusing at different rates and the only visible part being the ash. In this section we will review the data from the on-board sensors showing that the UAVs interacted with plumes on a number of occasions, and assess the accuracy of the hypothesis. to errors in predictable ways as indicated in Table 6.
The responses in Table 6 equate to a required increase or decrease in lift, compensating for a decrease or increase in lift, respectively. There are a number of parameters that can effect lift generation, the most effective being airspeed.
Turbulent air was expected upon entering the plume, with a higher density than ambient air. The turbulence felt in the plume is caused by the circulation of hotter air rising to the top of the plume and cooling, before sinking down again. The average temperature in the plume must be equivalent to the ambient temperature, else buoyancy would not be neutral. The centre of the plume is expected to be roughly equivalent to ambient air temperature, whereas the bottom is expected to be warmer and the top is expected to be cooler.
An increase in the activity and magnitude of Z-axis acceleration indicates turbulence, with increased high-frequency activity in the pitch data too. A discrete increase in air density would see a response from the TECS controller, equivalent to being above the target altitude. Changes in wind speed also cause a response, however they tend to be gradual therefore causing nondiscrete responses.
This section will consider each flight in turn, discussing the flight in general and, more specifically, the detection of the volcanic plume during the mission. Data collected by sensors on-board each flight will be considered and compared. It should be noted that the Skywalker flown for Flight B and Flight C was fitted with a motor that was not suited to the mission profile flown. The cruise throttle was near 100%, leaving little excess thrust for efficient climbing. The motor was replaced before Flight D, hence the significant change in cruising throttle setting.

| Flight A
The camera (Figure 7) showed The turbulence experienced by the Zephyr II in the plume was 1.9 km from the summit and was easily distinguishable in the logs.
It decided after this flight that Temp/RH sensors would be beneficial for further identification of the plume, only adding a small mass to the system. Real-time processing of this turbulence data can be moved on-board, however alone it will not suffice as a method for sensing the plume because turbulence can also be encountered in nonplume environments, such as clouds.   Temp/RH data ( Figure 14) show some differences between the iMet and the AMP. Outside this plot the iMet RH saturates in the cloud on the ascent and descent, and registers 0% during the summitapproach leg, the AMP peaks at around 85% and is mostly nonzero during the approach leg. One explanation for the lack of saturation on the AMP is the large protective shield fitted for protection of the delicate sensors, as seen in Figure 5.

| Flight D
This mission varies from the others presented here most notably because there was a section of piloted flight, albeit with augmentation from the autopilot. Figure 7 clearly shows that the plume was flown through. Where previous flights intercepted the plume at least 1.8 km from the crater, this one was approximately 0.6 km from the crater. Before the plume contact the UAV overflew the crater and loitered for 4 min. Figure 17 shows data from the initial summit overflight until after the piloted section of flight (see Figure 16). Note the magnitude of the acceleration data in Figure 17; the vertical accelerations recorded were up to four times greater than anything previously flown through. This can be attributed to the state of the plume at this distance from the summit, as it is still visually turbulent and not yet at neutral buoyancy (i.e., it is still rising, albeit at a slower rate than after initial expulsion). The pitch data also indicate turbulence at the plume point, however little else can be gleaned from the pitch or throttle data for this flight due to the aircraft mode. Fully automated flight would have enabled a more thorough analysis of the plume from an aircraft control perspective.
The nonneutral buoyancy observed suggested that there would be an increase in temperature during the fly-through. Figure 18 shows a relatively consistent temperature of 3.7°C during loitering flight and an increase of approximately 1.1°C during the plume flythrough. The AMP Temp registered approximately 0.9°C higher than the iMet, but peaked only around 0.5°C in the plume. The differences could be explained by calibration errors, however the sensors agree that in this instance the in-plume temperature is higher than ambient temperature, and is approximately 5°C.
RH data showed similar responses to the temperature, with a consistent 64% during the loiter phase and an increase to 74% in the plume. The iMet has previously shown a 20% RH increase in the plume, but here it is only 10%. One explanation for this is the mounting location of the sensor, as for this flight it was located in the nose rather than on the outside of the fuselage. The iMet data respond with acceptable levels of delay when compared to the flight data turbulence, however further testing of response times at flight speed is needed to confirm the exact point at which RH increases relative to the Z-axis acceleration.

| In-plume-detection metric
As shown in Figures 12, 15, and 19, P InPlume generates a value that peaks at the identified plume points of the flights. In addition to the previously identified plume points there are some additional unity values shown, such as in Figure 19 where there are peaks at around 27 min. It is suggested that the metric presented in this paper is a good indicator of UAV/plume interaction, and that it should be used in real-time for operational decision making when targeting plume fly-throughs. Further testing could include the inclusion of CO 2 or SO 2 sensors, which should sense changes in the plume compared to in ambient air. These could be included provided they are capable of sensing the expected change, are suitably light, and have low power requirements.

| CONCLUSIONS AND FURTHER WORK
This paper has shown that fixed-wing UAVs can be used to collect small ash samples from the plume of Volcán de Fuego in Guatemala.
Flight B here proves the concept of airborne ash collection using SEM stubs. Given an appropriate collection mechanism, the aerial sampling of ash with a representative PSD from within a plume has been shown to be possible. These data shall be used to better model the effects of volcanic ash on aircraft and be input into active aviation management tools.
Vertical acceleration of the aircraft was combined with altitude, pressure data, and humidity data to identify when the UAV was in a plume. These data were then combined to form P InPlume , which reached a value of 1 when the aircraft was in a plume over the three applicable flights presented here. Additional sensors, such as gas sensors, could be added for the generation of P InPlume , provided they add to the robustness of the metric and are suitably small and lightweight.
The team succeeded in finding a modus operandi which results in successful UAV flights for monitoring the volcano. With proper automation and education, the levels of expertise required to carry out monitoring missions can be reduced such that a wider range of people and local agencies could use the technology developed here.
Future aims of this project include flying through the plume using an automated tracking algorithm that considers the wind speed and direction, the input being a real-time implementation of the in-plume- | 1209 detection metric developed in this paper. This would enable sampling ash at a proximal point, then as the plume moves downwind sampling again from the same point at predefined radial distances. The analysis of these samples would then give the rate at which ash falls out of the plume, an important characteristic for aviation management. The automation of this process should increase reliability and repeatability.
Additional further work includes the development of a real-time trajectory planner that uses on-board computing to find a suitable, near-optimal, path to the area of interest, taking into account obstacles in the airspace and aircraft flight performance. These flights could be made into long endurance missions by incorporating energy scavenging algorithms. BVLOS operations with multiple airborne UAVs are also of interest, possibly to collect different data types from the same plume. The conditions around the crater will be investigated by deploying single-use remote sensors from the aircraft which will send data back to the operator in real-time.

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section.