The use of UAV remote sensing for observing lava dome emplacement and areas of potential lahar hazards: An example from the 2017–2019 eruption crisis at Mount Agung in Bali

https://doi.org/10.1016/j.jvolgeores.2021.107255Get rights and content

Highlights

  • UAV images with image enhancement and machine learning increase DTM's accuracies.

  • Multi-temporal point clouds co-registration is critical for volumetric analysis.

  • Lava dome emplacements associate with eruptions and potential lahar hazards.

  • DTM difference can recognize lahar starting points and supplied volume accurately.

  • The lahar inundation simulation reveals the areas of potential lahar hazard.

Abstract

Mount Agung (the highest volcano in Bali, Indonesia) began to erupt on November 21, 2017, after having been dormant for 53 years. More than 100,000 people were evacuated within the hazard zone between September 2017 (when the highest volcanic alert was issued) and early 2018. The eruptions continued until June 2019, accompanied by at least 110 explosions. During the eruptive crisis, the observation of the lava dome's emplacement was essential for mitigating the potential hazard. Details of the lava dome growth, including the volumetric changes and effusion rates, provide valuable information about potential eruption scenarios and lahar depositions. In this paper, the essential role of multi-temporal unmanned aerial vehicle (UAV) images in the monitoring of Mt. Agung's lava dome, and in determining the areas of potential lahar hazards during the crisis between 2017 and 2019 is described. A fixed-wing UAV was launched outside the hazard zone to photograph the lava dome on five occasions. Image enhancement, machine learning, and photogrammetry were combined to improve image quality, remove point clouds outliers, and generate digital terrain models (DTMs) and orthoimages. The analysis of the obtained DTMs and orthoimages resulted in qualitative and quantitative data highlighting the changes inside the crater and on the surrounding slopes. These results reveal that, from November 25 to December 16, 2017, the lava dome grew vertically by 126 m and reached a volume of 26.86 ± 0.64 × 106 m3. In addition, its surface experienced a maximal uplift of approximately 52 m until July 2019 with the emergence of a new dome with a volume estimated at 9.52 ± 0.086 × 106 m3. The difference between the DTMs of 2017 and 2019 reveals the total volume of erupted material (886,100 ± 8000 m3) that was deposited on the surrounding slopes. According to the lahar inundation simulation, the deposited material may cause dangerous lahars in 21 drainages, which extend in the north (N), north-east (N-E), south (S), south-east (S-E), and south-west (S-W) sectors of the volcano. This paper presents the use of UAV remote sensing for the production of high-spatial resolution DTMs, which can be used to both observe the emplacement of a lava dome, and to identify areas with potential lahar risk during a volcano crisis.

Introduction

The highest peak on Bali Island (Indonesia), with an altitude of 3000 m above sea level, Mt. Agung poses a serious threat to more than 100,000 residents who live in the volcanic hazard zone, 9–12 km from the crater. The last recorded eruption in 1963 is one of the ten most devastating volcanic eruptions of the twentieth century (Self and Rampino, 2012), with a volcanic explosivity index of 4–5 (Fontijn et al., 2015); approximately 1000 people were killed by pyroclastic flows (Tanguy et al., 1998). The secondary volcanic hazard, lahars, reportedly wiped out the landscape and killed at least 200 people from the intense rainfall after the eruption (Zen and Hadikusumo, 1964). This hazard history attracted the attention of the Indonesian Center for Volcanology and Geological Hazard Mitigation (CVGHM) because the reawakening signs of Mt. Agung's volcanic activity were evident in late August 2017 after 53-year of dormancy. The signs of unrest began with an intense swarm of earthquakes located between Mt. Agung and Batur with anomalous gas emission and expanding of hot areas within the summit crater (Syahbana et al., 2019). Progressive increases in volcanic activity in September 2017 forced the CVGHM to raise the volcanic alert level to Level II on September 14; it was increased to Level III four days later. The rapidly increasing seismicity peaked on September 22 and prompted a change to the highest alert level (Level IV) on September 18. The timeline of the awakening and eruption events of Mt. Agung in 2017–2019 is presented in Fig. 1.

The highest volcanic alert level triggered the evacuation; this level indicates that hazardous eruptions may happen and affect the inhabitants (ESDM, 2011). The CVGHM recommended that the area within a radius of 9–12 km from the summit crater should be evacuated. The Indonesian National Agency for Disaster Management followed this recommendation by coordinating the evacuation of inhabitants within the hazard zone. However, the volcano did not erupt until November 21. Prior to this, the seismic event rates decreased sharply on October 20, and the alert level was lowered to Level III on October 29, 2017. The decrease in the seismic event rate remained stable for the next two weeks, until it began to increase again in mid-November. On November 21, 2017, the initial phreatomagmatic phases started, which was followed by the main magmatic eruptions four days later. This event was considered to be the beginning of the intensive explosions and effusions period. Finally, the alert level was raised to the highest level on November 26.

Mt. Agung underwent strong magmatic eruptions for at least two months following the first eruption. More than 30 magmatic explosions were recorded, and the highest ash column reached approximately 4000 m above the summit (the eruption column heights are indicated by gray circles in Fig. 1). These explosions exhibited several characteristic types of eruptions, i.e., phreatomagmatic (producing enormous black ash plumes and ejecting a significant juvenile magmatic component), magmatic lava effusion (producing lava effusion), and magmatic strombolian (expanding gases that eject incandescent lava), which could transition from one type to another or occur simultaneously during one particular event. The magmatic explosion events are depicted by brown vertical lines in Fig. 1.

During this two-months period, the low-frequency seismicity rate increased, culminating in a Strombolian-type explosion on January 19, 2018 (Syahbana et al., 2019). Thereafter, the explosion frequency gradually decreased, and the period of intensive explosions and effusions activity of Mt. Agung was considered to have ended. Consequently, the alert status was downgraded to Level III on February 10, 2018. Nevertheless, at this alert level, magmatic explosions were recorded until June 2019; at least 80 small explosions were registered, some of which produced 2–5 km high plumes over the summit. Another event involving Strombolian activity occurred on July 2, 2018, throwing incandescent material as far as 2–3 km from the summit. This activity was recorded as a series of seismic explosion signals. The extrusion of lava and the explosion frequency increased until July 27, 2018. Subsequently, the explosions decreased in magnitude and frequency over the next five months.

Another series of magmatic eruptions started on December 20, 2018, with explosion columns reaching 200–5500 m above the summit. The explosions continued until June 2019, including several small-to-medium explosions, lava effusion, and continuous emission of ash in various periods. The last eruption of the 2017–2019 eruptive crisis at Mt. Agung occurred on June 13, 2019. However, volcano-tectonic (VT) seismicity was still detected the next year with decreasing magnitude and frequency. Furthermore, the ash analysis of the phreatomagmatic deposits below the vent showed the existence of juvenile material and fragments of lava from the 1963 eruption (ESDM, 2017).

Lava dome eruptions are typically characterized by explosions, lava effusion and growth, dome collapses, and even cooling stages (Herd et al., 2005; Darmawan et al., 2018a). Understanding lava dome changes during an eruptive crisis is essential to mitigating hazards because they reflect the magmatic pressurization (Pallister et al., 2013), the energy of an eruption (Wadge et al., 2006), and the impact of future eruptive activity (Diefenbach et al., 2012). The estimation of magma extrusion rates from vents or the quantification of material deposition is vital data (Schilling et al., 2008; Kubanek et al., 2015). Furthermore, it is fundamental to measure topographical changes, including volumetric variations of the dome or lava flow because of its close relation to the potential hazard.

Frequent monitoring is required to observe lava emplacement, particularly during intense volcanic activity. The primary required data includes a digital terrain model (DTM) because it represents the earth surface and can be used to analyze the quantitative changes during sequential dates. However, obtaining high-spatial and high-temporal-resolution DTMs at dome sites presents difficulties typically due to restricted access, hazardous conditions, high altitudes, and extreme environmental conditions of lava domes. Satellite remote sensing techniques, such as radar interferometry (e.g., the SRTM mission and the TanDEM-X satellite) and stereo-optical satellite sensors (e.g., Pleiades, ASTER, and ALOS), have been widely used for volcanological studies (Poland et al., 2020) because they provide high temporal resolution and large coverage area at several scales and resolutions. However, the imaging geometry of radar systems causes distortions, such as shadowing and layover of high relief terrain (Deng et al., 2019). The conventional use of repeat-pass SAR-interferometry (InSAR) for monitoring active volcanoes can fail during the interferometric process, due to the presence of decorrelation effects (Kubanek et al., 2015). Furthermore, stereo-optical satellite sensors are limited by cloud cover, particularly in tropical areas, and the unavailability of observations for specific dates and locations (Bonali et al., 2019). Other studies using airborne light detection and ranging (LiDAR) or combined airborne hyperspectral technology with LiDAR can also present high-spectral-resolution images with high-density 3D point clouds from laser return signals (Favalli et al., 2010; Behncke et al., 2016; Kereszturi et al., 2018). Unfortunately, these methods are expensive and difficult to apply during eruptive events; in addition, flying around an erupting volcano is dangerous.

An alternative method without these deficiencies is the use of an unmanned aerial vehicle (UAV) equipped with a digital camera for observing the active lava dome. We collected images over the summit during different periods (during the pre-eruption, during the period of intensive explosions and effusions activity, and after the intermittent eruption) and applied the structure-from-motion (SfM) method to perform photogrammetric aerial triangulation (bundle adjustment) and generate 3D dense point clouds with the multi-view stereo (MVS) algorithm. The SfM technique is used to reconstruct the 3D structure of objects from image sequences and calculate both interior and exterior orientations simultaneously (Carrivick et al., 2016). Combining this technique with a UAV system can produce high-density point clouds to generate DTMs and orthoimages (Nakano et al., 2014; De Beni et al., 2019). The MVS generates a natural feature surface digital surface model (DSM), which must be filtered into a DTM to minimize the chances of falsely calculating the surface changes affected by land cover elevation, such as trees or buildings. In this study, we only investigated the topographic and volumetric changes within a bare land area, so that the DSM is considered equivalent to the DTM.

Advances in computer vision enable the use of photogrammetric techniques to generate 3D point clouds, high-resolution DTMs and orthoimages from UAV images. UAV photogrammetry has been applied to volcano monitoring not only for investigating topographic and structural changes on an active volcano (Nakano et al., 2014; Darmawan et al., 2018a; De Beni et al., 2019) but also for measuring deformation (Favalli et al., 2010; Bonali et al., 2019), observing structural weakening (Darmawan et al., 2018b), detecting geometric changes and fractures (Muller et al., 2017), detecting volcanic gases (Pieri et al., 2013), and estimating the volume of material destroyed by an explosion (Arámbula-Mendoza et al., 2020). Furthermore, terrestrial photogrammetry has been applied for studying volcanoes, using photographs captured with cameras located at fixed positions surrounding the crater. Major et al. (2009) used close-range images from remote areas to identify changes of the lava dome of Mount St. Helens. Furthermore, Carr et al. (2019) described how SfM-MVS was used to generate DTMs over a period of 14 days when Mt. Sinabung experienced a large eruption. The DTMs were used to measure the lava dome volume, the instabilities, and the rates of discharge and emplacement. Another comprehensive study was conducted by Deng et al. (2019) who used a combination of terrestrial photogrammetry and radar interferometry to generate new DTM data without data gaps, which improved the accuracy of simulations of pyroclastic flows and lahars, and their associated inundation zones. The results showed that accurately and frequently updated DTMs were crucial to volcanic-flow simulation, including pyroclastic and lahar flows.

This paper presents UAV remote sensing for monitoring lava dome emplacements and topographical changes for determining the areas of potential lahar hazards during the 2017–2019 eruptive crisis at Mt. Agung in Bali, Indonesia. The UAV missions allowed the investigation of the active lava dome while operating outside of the hazard zone; the UAV could not fly near the summit owing to the hazardous conditions. The 2017–2019 Mt. Agung eruptions deposited ash on its upper flanks mainly during the period of large and intense ash emissions (e.g., between November 21, 2017, and January 24, 2018, and between June 28 and July 27, 2018), as depicted in the timeline in Fig. 1. These ash deposits and other volcanic debris could easily be transported by the run-off from intense rainfall; it generated lahars, which are the most threatening hazard after an eruption. More than 30 lahar events had occurred between 2017 and the present (August 2020), thereby destroying roads, bridges, and other infrastructure along their path. In this study, the amount of material deposited on the slopes was determined, and the topographical changes resulting from the 2017–2019 eruptions were analyzed with a UAV campaign before and after eruption phase. Moreover, the probable locations for the initiation of lahars with possible lahar volumes were determined, which allowed the prediction of lahar flow inundation zones. The results of the lahar flow simulations were used to identify potential lahar hazard in the downstream area. In this paper, the performance and benefits of obtaining data using UAV-remote sensing technologies for the observation of active lava dome changes and the identification of areas of potential lahar hazards in the future are discussed.

Section snippets

UAV survey campaign

The multi-temporal UAV campaigns were conducted on five dates between 2017 and 2019 (Table 1). The lava dome was observed on three days during the pre-eruption stage (October 19–21, 2017; the days are named “T1”, “T2”, and “T3”, respectively), during the intensive explosions and effusions event (on December 16, 2017; “T4”), and after intermittent eruption event (on July 6, 2019; “T5”). Acquiring UAV images of the lava dome site was challenging because of the extreme environmental conditions,

Results

3D point clouds, DTMs, and orthoimages were generated from the five dates of the UAV campaign using SfM-MVS photogrammetric techniques. The details of the original data are summarized in Table 1. The data covered areas from 3.51 to 33.25 km2 (Table 2), including the entire 0.67 km2 of the crater. All sides of the lava dome were well covered, except for dates T3 and T4, because the cloud was covering some parts of the dome (Fig. 4d and f). We demonstrated the effectiveness of data pre-processing

Discussion

The results highlight the feasibility of UAV-based remote sensing for analyzing lava dome changes at Mt. Agung during the eruptive period of 2017–2019. The data pre-processing consisted of three steps: image enhancement, outlier removal, and multi-temporal point clouds co-registration. The methodology significantly reduced the geometrical errors, increased the reliability, and achieved less than 1 m of point clouds co-registration error. For the estimation of volumetric errors on UAV-derived

Conclusions

UAV-based remote sensing was successfully applied for lava dome monitoring and lahar hazard modeling. The results of this study demonstrate that data pre-processing of UAV images improves accuracy and enables a quantitative description of lava dome changes. A campaign using a fixed-wing UAV is advantageous because of its flexible deployment, low risk, and ability to withstand long-endurance missions. In all missions, the UAVs were able to fly 300–1100 m above the lava dome; they traveled more

Declaration of Competing Interest

None.

Acknowledgment

We would like to thank the Indonesian National Disaster Management Authority for giving the authority to operate the UAV campaign during the eruption periods, and the Disaster Response Unit - Gadjah Mada University for supporting the funding in the first data acquisition. We would also like to thank the Indonesian Center for Volcanology and Geological Hazard Mitigation and Aeroterrascan for providing additional UAV images in this research. Additionally, we acknowledge Wahyu Widiyanto, Rio Andi,

References (59)

  • J.J. Major et al.

    Monitoring lava-dome growth during the 2004-2008 Mount St. Helens, Washington, eruption using oblique terrestrial photography

    Earth Planet. Sci. Lett.

    (2009)
  • J.-C. Padró et al.

    Comparison of four UAV georeferencing methods for environmental monitoring purposes focusing on the combined use with airborne and satellite remote sensing platforms

    Int. J. Appl. Earth Obs. Geoinf.

    (2019)
  • J.S. Pallister et al.

    Merapi 2010 eruption—Chronology and extrusion rates monitored with satellite radar and used in eruption forecasting

    J. Volcanol. Geotherm. Res.

    (2013)
  • G. Wadge et al.

    The magma budget of Volcán Arenal, Costa Rica from 1980 to 2004

    J. Volcanol. Geotherm. Res.

    (2006)
  • P. Agrafiotis et al.

    Underwater image enhancement before three-dimensional (3D) reconstruction and orthoimage production steps_ is it worth?

  • S.A. Alasal et al.

    Improving passive 3d model reconstruction using image enhancement

  • A. Al-Rawabdeh et al.

    A robust registration algorithm for point clouds from uav images for change detection

  • R. Arámbula-Mendoza et al.

    Destruction of a lava dome observed with photogrammetry, acoustic and seismic sensors at Volcán de Colima, Mexico

    J. Volcanol. Geotherm. Res.

    (2020)
  • D.W.D. Arnold et al.

    Decaying lava extrusion rate at El Reventador volcano, Ecuador, measured using high-resolution satellite radar

    J. Geophys. Res. Solid Earth

    (2017)
  • A. Ballabeni et al.

    Advances in image pre-processing to improve automated 3d reconstruction

  • B. Behncke et al.

    Lidar surveys reveal eruptive volumes and rates at Etna, 2007-2010

    Geophys. Res. Lett.

    (2016)
  • P.J. Besl et al.

    A method for registration of 3-D shapes

    IEEE Trans. Pattern Anal. Mach. Intell.

    (1992)
  • L. Breiman

    Random Forests

    Mach. Learn.

    (2001)
  • I.S.G. Campos et al.

    A height estimation approach for terrain following flights from monocular vision

    Sensors

    (2016)
  • J. Carrivick et al.

    Structure from Motion in the Geosciences

    (2016)
  • H. Darmawan et al.

    Structural weakening of the Merapi dome identified by drone photogrammetry after the 2010 eruption

    Nat. Hazards Earth Syst. Sci.

    (2018)
  • F. Deng et al.

    High-resolution DEM generation from spaceborne and terrestrial remote sensing data for improved volcano hazard assessment - a case study at Nevado del Ruiz, Colombia

    Remote Sens. Environ.

    (2019)
  • A. Diefenbach et al.

    Rapid, low-cost photogrammetry to monitor volcanic eruptions: an example from Mount St. Helens, Washington, USA

    Bull. Volcanol.

    (2012)
  • ESDM

    Minister of Energy and Mineral Resources Regulation Number 15 year 2011

    (2011)
  • Cited by (13)

    • Determination of potential secondary lahar hazard areas based on pre-and post-eruption UAV DEMs: Automatic identification of initial lahar starting points and supplied lahar volume

      2022, International Journal of Applied Earth Observation and Geoinformation
      Citation Excerpt :

      However, since the pre-eruption DEM (T1) was generated by UAV images with only two flight lines in the southeast direction, thus, the T1 DEM only covers the southeast of the volcano (the flight trajectory and coverage area are depicted in Fig. 3). Further details of the co-registration strategy of the datasets, photogrammetric processing, and quality assessments of the UAV products can be found in the papers Andaru et al. (2021); Andaru et al., (2022). The ideal datasets for estimating actual LSP and SLV are the DEMs of pre-and post-eruptions.

    • Unmanned aerial vehicles (UAVs) as a tool for hazard assessment: The 2021 eruption of Cumbre Vieja volcano, La Palma Island (Spain)

      2022, Science of the Total Environment
      Citation Excerpt :

      Volcanic eruptions are hazardous and particularly challenging for ground-data collection or direct observations (Walter et al., 2020), and require remote sensing tools for their monitoring and risk assessment. However, applicability of satellite remote sensing in the study of volcanic events remains limited because of the reliance on optimal meteorological conditions and the consequent unavailability of observations for specific dates and locations (Andaru et al., 2021; Bonali et al., 2019), with the exception of those equipped with thermal and radar sensors (e.g. SAR), which can work at night, with cloud cover, etc. In addition, despite the advances in satellite remote sensing technology, it is still difficult to monitor these dynamic phenomena due to their coarse spatial image resolution and their long re-acquisition intervals (Turner et al., 2017).

    View all citing articles on Scopus
    View full text