**2. Background**

Fuego (alt. 3763 m; 14.4◦ N, 90.8◦ W) is the southernmost and youngest of five vents comprising the Fuego-Acatenango massif, located on the Central American volcanic arc (CAVA) in Guatemala (Figure 1). The CAVA is a chain of Quaternary stratovolcanoes linked to subduction of the Cocos plate under the Caribbean plate [29,30]. Fuego is one of the most active volcanoes in Central America, with over 60 recorded eruptions since 1524 [31]. Most of Fuego's historic eruptions have been classified with a Volcanic Explosivity Index (VEI) of 2 or 3, interspersed with several larger VEI 4 events, the most recent of which occurred in 1974 [31–34]. The region surrounding Fuego is highly populated, with ~9000 people living within the high hazard zone potentially threatened by PDCs, and more than one million within 30 km of the volcano. Larger eruptions of Fuego also potentially threaten the national capital (Guatemala City) and main international airport, which are only ~40 km away. During the 3 June 2018 eruption of Fuego the ash emissions forced the closure of the La Aurora International Airport in Guatemala City.

Fuego undergoes long periods of relatively low background activity, interrupted by periods of intense, or 'paroxysmal', activity that consists of lava flows, lava fountains and PDCs [33,34]. The latter are typically channelled down one or more of the barrancas (valleys) on the flanks of the volcano (Figure 1b), and can be a significant hazard, either by directly impacting populated areas or by providing material for lahars during the rainy season [32,35]. This broad range of eruptive styles makes Fuego's activity both scientifically interesting and challenging to forecast. Furthermore, the persistent activity results in constant morphological variation in the vent region and on the flanks (barrancas), which is difficult to track using ground-based observations.

Fuego was very active in 2018, with 3 paroxysmal eruptions in February, June, and November. We focus here on the 31 January–2 February 2018 eruption, a Strombolian event of VEI 2 (Figure 2) that was well-captured in PS data. The June 2018 eruption, which resulted in extensive loss of life and damage to infrastructure, involved a more complex sequence of events, and will be the subject of a separate paper. The 31 January–2 February 2018 eruption started with continuous lava fountains and two lava flows, directed NW from the summit. Subsequently, PDCs formed on the east flank and deposited material in two of the seven barrancas which surround the edifice. A significant amount of ash was also emitted and dispersed predominantly to the west by prevailing winds. The activity was also monitored by webcams, providing a visible record of the sequence of events. PS images collected before, during and after the eruption provide a novel, detailed perspective on the volcanic activity for monitoring and mapping the distribution of the volcanic deposits.

**Figure 2.** Timeline of the February 2018 eruption of Fuego and its main physical and volcanological features. Data sources: (**a**) Smithsonian Institution Global Volcanism Program (GVP) [24]; Planet Labs data coverage [26]; (**b**) MODVOLC website [36]; (**c**) VAAC database [37]. Planet labs imagery timeline is in Table S1.

#### **3. Data and Methods**

#### *3.1. The Planet Labs CubeSat Constellation*

The Planet Labs (PL) CubeSat constellation comprises (as of early 2019) over 150 CubeSats or 'Doves', each roughly 10 × 10 × 30 cm in size (i.e., three-unit or 3U CubeSats), orbiting in two near-polar, Sun-synchronous (SS) orbits with ~8◦ and ~98◦ inclination at an altitude of ~475 km. Each Dove carries a telescope and a 6600 × 4400 pixel CCD array, which acquires both visible (red–green–blue or RGB) and near-infrared (NIR) PS data with 12-bit radiometric resolution. PS scenes are frame images (i.e., an entire scene is imaged by the CCD at an instant in time, unlike pushbroom sensors) with approximate dimensions of 25–30 × 8–10 km, and a native spatial resolution of 3.7 m, resampled to 3 m for delivery. The Doves have been deployed since 2014 in 'flocks' from various launch vehicles. Early CubeSats (operational in 2014–2017) were released from the International Space Station (ISS) into the ISS orbit with 52◦ inclination at ~375 km altitude, whereas more recent PS data (including the data used in this study) are acquired by the near-polar SS-orbiting flocks (Table 2). The Dove constellation images the entire Earth landmass on a daily basis, with overpass times in the local morning hours. To achieve daily contiguous global coverage, the swaths of each consecutive Dove overlap in the across-track direction, providing the opportunity for multiple scene acquisitions within a few minutes in the overlap region [35]. Planet Labs also operates a fleet of five RapidEye satellites and 14 SkySats, which provide even higher spatial resolution.


**Table 2.** Specifications of Planet Labs deployments in International Space Station (ISS) and Sun synchronous orbits (SSO).

Several PS data products with different processing levels are available; here we use the 'Analytic\_SR' data products which are 16-bit calibrated orthorectified surface reflectance data with a positional accuracy of better than 10 m.

#### *3.2. Data Processing*

This study of the February 2018 activity at Fuego is based on change detection techniques-a comparison of images pre, syn and post-eruption of the same area at different times. The activity at Fuego permits the evaluation of PS data for a variety of eruption products including lava flows, trephra fall deposits and PDCs.

PS imagery was browsed, and products downloaded using the Planet Explorer interface [26]. Images from before and after the 31 January–2 February eruption were identified, inspected for cloud cover and clarity, and downloaded. Visual inspection of the images was performed by selecting suitable RGB colour stretching values to highlight volcanic deposits and structures. Images were merged and cropped to the extent of the area of interest. Visual comparison of pre- and post-eruption images was performed to identify and map changes, including new deposits produced by tephra fall, lava flows and PDCs. Visual comparisons primarily used the visible bands, but the NIR band was also used in some cases (substituting for the green band in the RGB composite images), particularly when changes in vegetation cover due to the volcanic activity were involved. Where new eruptive material was deposited over areas previously covered by vegetation, a simple change detection strategy based on the

NDVI difference was applied. NDVI values were calculated for each image from the surface reflectance values, corresponding to the red and NIR bands [38,39]. We then subtracted the NDVI values in the pre-eruption images from the post-eruption NDVIs, for spatially collocated pixels, yielding a raster dataset of pre/post-eruption NDVI differences. A negative NDVI difference thus indicated areas where vegetation has been damaged and/or covered by volcanic deposits; the NDVI difference can have any value between −1 and 1, from completely unvegetated (or ash-covered) to completely vegetated (or ash-free). To simplify the interpretation a threshold was chosen, usually −0.1 or −0.125, to represent the change that was considered significant and interpreted as being produced unequivocally by the deposition of volcanic material (e.g., tephra fall). Cloud cover limits the use of reflectance based remote sensing data in the visible and NIR bands. Clouds and cloud shadows were visually identified and masked in all the images; such masks were then applied to exclude cloudy areas and any associated noise and biases from the data analysis. Finally, the raster maps obtained from the NDVI differences were simplified, by applying a low-pass filter (e.g., 15 × 15 neighbour pixel average operation) to the threshold maps and converting them to polygons that mapped the tephra-covered areas.

The NDVI difference approach in general is not suitable for detecting changes in areas that were originally non-vegetated, like the upper flanks of the volcano or the active barrancas. In such cases other band combinations can be tested to see if a suitable band or combination of bands could be used to detect and identify the new deposits. In particular, the visible and NIR bands of a pre-eruption image were subtracted from the corresponding band of the post-eruption image. A trial-and-error approach was adopted using single bands (e.g., pre-eruption Band 1 minus pre-eruption Band 1, etc.) but results were not satisfactory. Also, the same band on pre and post-eruption dates was analyzed with the same approach as NDVI differences for a time-dependent analysis.

Processing of the visible, NDVI and other band combinations yielded a series of maps of changes between the pre- and post-eruption images, which were then interpreted in terms of new eruptive deposits and other volcanic processes. Additionally, volcanic structures in the vent region and upper flanks of Fuego were also mapped visually using RGB images. PDCs can be highly erosive; they engrave channels and incorporate pre-existing material as they move downhill. This produces characteristic features along the PDC path that can be detected by comparing pre- and post-eruption images. Comparing images from a longer time-series with those captured during short-livedevents allow us to distinguish changes in crater morphology associated with background and paroxysmal activity.

#### **4. Results**

Eruptive deposits and volcanic structures mapped with the methods previously described were compiled into a GIS platform (ArcGis,® Environmental Systems Research Institute, Redlands, California, United States) for interpretations and analysis. Here we describe the different types of deposits and structures separately as interpreted from our analysis.

#### *4.1. Lava Flows*

Deposits optically identified and interpreted as two lava flows erupted during the February 2018 paroxysm are shown in Figure 3. These lava flows moved down the NW slope of Fuego. The longest of the two reaches a length of ~2000 m and they respectively cover areas of 0.12 and 0.18 km2. They are representative of the beginning of the activity; a Strombolian style eruption characterized by sustained lava fountains and bombs. At the beginning of the activity the upper flank is characterized by spatter deposits and lava flows (Figure 3a). Distinguishing these deposits is difficult because of the lack of contrast between them. In this case, the morphology plays a crucial role for the distinction between the smoother lava flows and the rugged surface of the spatter deposit. Automated detection of different deposits on the highest part of the edifice is more difficult. In fact, neither the NDVI difference nor any other simpler band operation gives good results for this kind of deposits. It is still possible to map the darker material using single bands and appropriate thresholds. Figure 3b shows band-2 pixel values with digital numbers (DN) <3000 (in a 16-bit image) in the vent region; similar results were obtained

for the other three bands. The result is closer to the optical map of the deposits shown in Figure 3a, but with some additional signal (likely due to tephra fall) on the upper west flank.

**Figure 3.** PlanetScope image of Fuego on February 2, 2018; 15:57 UTC. (**a**) Lava flows and spatter deposits optically mapped are shown in orange and yellow respectively. (**b**) Result for single Band 2, using a threshold DN < 3000, for comparison.
