Detailed multidisciplinary monitoring reveals pre- and co-eruptive signals at Nyamulagira volcano (North Kivu, Democratic Republic of Congo)

Abstract

This paper presents a thorough description of Nyamulagira’s January 2010 volcanic eruption (North Kivu, Democratic Republic of Congo), based on a combination of field observation and ground-based and space-borne data. It is the first eruption in the Virunga Volcanic Province that has been described by a combination of several modern monitoring techniques. The 2010 eruption lasted 26 days and emitted ∼45.5 × 10⁶ m³ of lava. Field observations divided the event into four eruptive stages delimited by major changes in effusive activity. These stages are consistent with those described by Pouclet (1976) for historical eruptions of Nyamulagira. Co-eruptive signals from ground deformation, seismicity, SO₂ emission and thermal flux correlate with the eruptive stages. Unambiguous pre-eruptive ground deformation was observed 3 weeks before the lava outburst, coinciding with a small but clear increase in the short period seismicity and SO₂ emission. The 3 weeks of precursors contrasts with the only precursory signal previously recognized in the Virunga Volcanic Province, the short-term increase of tremor and long period seismicity, which, for example, were only detected less than 2 h prior to the 2010 eruption. The present paper is the most detailed picture of a typical flank eruption of this volcano. It provides valuable tools for re-examining former—mostly qualitative—descriptions of historical Nyamulagira eruptions that occurred during the colonial period.

Appendices: Monitoring methods

Appendix 1: Remote sensing support for field observation

During the first days of the eruption, bad visibility due to difficult weather conditions prevented the accurate description of the eruptive activity from helicopter flights. Satellite imagery offered invaluable tools to detect and locate eruptive activity. We used ENVISAT-ASAR images to locate the flank vent (Fig. 21). Coherence images resulting from the production of SAR interferograms allowed the detection of lava flow and tephra deposits in the caldera, as well as lava flow paths on the volcano flanks (Fig. 22). ASTER TIR bands and contrasts of radiance on lava flows helped detect the active lava flow branches during some eruption stages. The compilation of Terra-ASTER, Landsat 5TM, Landsat 7ETM+ and EO1-ALI images allowed the fully mapping of the lava flows and fissures with a spatial resolution of 15 to 30 m. We used this mapping to estimate lava flow length, surface and volume. The volume of lava emitted during the 2010 eruption was obtained by multiplying the mapped lava flow surface by the mean lava flow thickness commonly adopted for Nyamulagira (i.e. 3 m) in previous publications (Smets et al. 2010).

Fig. 21

Comparison of ENVISAT-ASAR amplitude images before (a) and after (b) the eruption onset. a and b correspond to images resulting from the average of five amplitude images. The amplitude image acquired on 8 January 2010 presented in b provided the team with the first accurate location of the flank vent and the filling of the pit crater by lava (red circles)

Fig. 22

SAR coherence images from interferograms covering time periods before the eruption (a) and during the first week of the event (b). c is a RGB-composite image where the red band is the coherence image of (b) and the green and blue bands are the coherence image of a. The comparison between a and b allowed for detection of the lava flow and tephra deposits in the caldera as well as one of the main branches of the flank lava flow

Appendix 2: InSAR monitoring

InSAR is now widely used to study a broad range of deformation of volcanic and seismic origins (e.g. Bürgmann et al. 2000; Lu et al. 2005; Wright et al. 2006; Yun et al. 2006; Simons and Rosen 2007; Fournier et al. 2010 and references therein).

Since 2005, systematic SAR acquisitions have been programmed for monitoring Nyamulagira and Nyiragongo volcanoes (e.g. d’Oreye et al. 2008; van Overbeke et al. 2010). SAR images acquired from various geometries (ascending and descending tracks, various look angles and wavelengths) and sensors on board of different satellites (ERS 1–2, JERS, ENVISAT, ALOS, RADARSAT 1–2, TERRASAR-X) have provided hundreds of images since 1996.

Specific tools for automatic InSAR processing were built to cope with the rapidly increasing database (d’Oreye and Celli 2010). Interferograms are processed using DORIS (Kampes and Usai 1999; Kampes et al. 2004), ROI-PAC (Rosen et al. 2004) or GAMMA (Wegmuller and Werner 1997) software. Deformation and coherence maps are usually available in common GIS format, typically within 30 min after reception of a new SAR scene.

In addition to classical SAR differential interferograms, we used the new MSBAS method developed by Samsonov and d’Oreye (2012) to generate 2D time-series (vertical and W–E) of ground deformation, simultaneously combining all the images captured by various satellites under multiple geometries. In the present work, we make use of 181 SAR images acquired under eight different geometries by three satellites. From these, we computed more than 1,000 interferograms covering the period from January 2003 to September 2010.

The method was evaluated with theoretical tests (Samsonov and d’Oreye 2012) and successfully applied for mapping natural ground deformation in Virunga Volcanic Province (Samsonov and d’Oreye 2012) and Campi Flegrei-Vesuvius (Italy) (Samsonov et al. 2014b) and anthropogenic ground deformation caused by mining along the French–German border (Samsonov et al. 2013a) and ground water extraction in the southern Saskatchewan, Canada (Samsonov et al. 2013b) and Bologna, Italy (Samsonov et al. 2014a) regions.

The MSBAS method has four main advantages: (a) It achieves combined temporal coverage over an extended period of time when data from many different sensors with different time periods are available; (b) the temporal resolution of the time series produced increases because it includes the combined sampling from all data sets, which helps to observe signals in greater detail and also to improve the quality of post-processing (i.e. in the case of filtering); (c) two or three components of the ground deformation vector are computed, which helps in the interpretation of observed ground deformation and further modelling and inversion, and (d) various sources of noise (i.e. tropospheric, ionospheric, topographic, orbital, thermal etc.) are averaged out during the processing improving the signal-to-noise ratio. The MSBAS time-series computed here provided us a picture of the vertical and W–E displacements, at unprecedented detail, before, during and after Nyamulagira’s eruption.

Appendix 3: GNSS monitoring network and data processing method

Each GNSS station is equipped with a Leica GMX 902 GG or GR10 receiver and a Leica AX1202 or AS10 antenna fixed on a 2.5-m high concrete monument. These stations track both GPS and GLONASS satellites. Signals from the stations, sampled at 1 or 15 s, are radioed in real time to the GVO using 900-MHz radios manufactured by Intuicom. When direct line of sight between stations and GVO is not possible, no more than one repeater relays the signal. Raw data are stored and pre-processed at the GVO and automatically transmitted by Internet to the National Museum of Natural History in Luxembourg. The preliminary estimation of the baseline vectors is calculated in real time at the GVO.

The present work makes use of GPS and GLONASS data from Bulengo, Rusayo and Goma stations (BLG, RSY and GVO). Kibati and Kibumba stations (KBB, KBT) were unavailable before January 18 because lightening damaged that segment of the network. Tsubi, Rubavu and Bobandana (TSI, RBV and BOB) were all installed after the eruption. We used only one reference station (NURK, installed 100 km to the East in Kigali, Rwanda), as other available references either suffer from data interruptions during the studied time period, have no official coordinates assigned and are not used in the IGS weekly solutions, or are located more than 1,000 km away from the study area.

GNSS data presented here were processed using Bernese 5.0 software (Dach et al. 2007). First, a linear adjustment of the reference station (Fig. 23) is computed using the weekly solutions from 1,521 to 1,565 GPS weeks in order to estimate a set of coordinates and their rate of change. Residuals are below 3–4 mm. GNSS data from each station are down-sampled to 30 s jointly with precise ephemerides and Earth observation parameters from CODE. A cutoff angle of 10° and the FES2004 model for oceanic load corrections have been used (Lyard 2006).

Fig. 23

Linear adjustment of coordinates for NURK station (left) and residuals (right)

Three types of processing were carried out:

• 1. Daily processing using precise point positioning in order to obtain good approximate coordinates for the network stations, as recommended in Dach (2007)

• 2. Daily network solutions considering reference station NURK as fixed. The computations were carried out using the Bernese Processing Engine, and the Quasi-Ionosphere-Free ambiguity resolution strategy was used (Dach 2007). The series of daily solutions shows 10–20-mm noise mainly due to use of only one fixed point for the adjustment. In spite of the a posteriori error estimations of the adjusted coordinates of 1–2 mm, it is known that Bernese underestimates the magnitude of the errors by factors of up to 23 times (Kashani et al. 2004).

• 3. Weekly (in terms of GPS week) network solutions by combination of the daily solutions. These computations were carried out in order to minimise the noise in the daily solutions. The combinations were done with the ADDNEQ2 utility of Bernese, using the normal equations results from the previous daily solutions. The repeatability of the coordinates in each weekly adjustment is of the order of 5–10 mm in each coordinate (east, north, up), so the quality of the results can be seen as EUREF class B stations (Adam et al. 1999).

The regional tendency produced by tectonic plate motion was suppressed by applying the rate displacements computed for NURK to the solutions in the linear adjustment of it coordinates. Computed with the theoretical model NUVEL1A, the rates for NURK and for the area of study only differ by 0.08 mm/year, so the application of the rates computed for NURK to the stations of the working network does not reflect any loss of generality.

Appendix 4: Tiltmeter monitoring network

Tiltmeter stations (Fig. 2) are equipped with analogue biaxial Applied Geomechanics models of the ‘AGI700’ series. Their resolution is better than 0.1 microradian. Data are digitised using a 13-bit A/D converter and sampled at 1 min. They are stored on site using CR100 data loggers manufactured by Campbell Scientific. Their download is realized on site using a laptop computer, and a pre-processing is performed at the GVO using the TSoft software (Van Camp and Vauterin 2005). Archives are stored at GVO and a copy is automatically transferred to the National Museum of Natural History in Luxembourg. Tiltmeters in Rusayo and Bulengo stations are installed on seismic pillars anchored in the bedrock and decoupled from the hosting building. In Ngangi and Munigi, instruments are installed on vertical stainless steel rods cemented in the lava, respectively, at 1–2 m depth in a lava tube and in an artificial vault dug into the ground.

Appendix 5: Ground-based SO₂ emission monitoring

The dynamics of gas-magma segregation and transport is recognized as one of the driving mechanisms for magma ascent at shallow depths. Volcanic plumes originated in this process represent by themselves a hazard to people and environment (e.g. Symonds et al. 1994; Oppenheimer 2003). In the case of Nyamulagira, one of the most conspicuous features of its volcanic activity is the magnitude of its gaseous emissions during eruptive periods compared to non-eruptive periods. Between eruptions, emissions seem to be close to zero or at least not detectable by the current monitoring techniques. For these reasons, monitoring the rate of outgassing (or flux in kilograms per second) is important to understanding the conditions of this volcano. Documented records of the style, speciation and magnitude of magma outgassing from Nyamulagira exist in the scientific literature from the 1980s (Aoki et al. 1985; Krueger et al. 1996; Carn and Bluth 2003; Bluth and Carn 2008; Chakrabarti et al. 2009). Due to logistical constraints, these studies were based on geochemical analysis of in situ sampling from sporadic field campaigns or on observations by satellite sensors.

To strengthen the gas monitoring capabilities of Nyamulagira and Nyiragongo, Chalmers University (Sweden), in cooperation with the GVO, installed one automatic remote sensing instrument in March 2004 to measure the gas flux of SO₂. This instrument is a single-beam scanning DOAS system (Galle et al. 2003; Johansson et al. 2009) acquiring spectra in the UV spectral region of radiation scattered in the atmosphere. DOAS station is located at the GVO’s Rusayo seismic station, about 18.7 km S of Nyamulagira’s crater. Solar panels and batteries provide power. The data are transmitted by a radio-link to the GVO, where they are evaluated and uploaded to the archive of the NOVAC project. Combined with plume velocity information, this instrument provides near-to-real-time SO₂ flux and plume dispersion characteristics (height, direction, width) with a typical time resolution of 10 min during sunlight hours. In 2005, the instrument was upgraded and incorporated as part of the NOVAC network. After the first installation at Virunga, four additional instruments have been deployed 17–24 km SW of the crater of Nyamulagira, at the sites Sake, Kunene, Buzi and Kingi. Due to logistical difficulties in accessing these stations, which were not operational before and during the eruption period, we only present the flux measurements from the Rusayo station in this study.

The DOAS method to retrieve gas fluxes from scanning measurements is well established, and the interested reader can found the details elsewhere (e.g. Galle et al. 2010). For the purpose of this paper, we only describe some considerations specific to the case under analysis.

Firstly, the spectral evaluation included a wavelength shift calibration based on correlation of selected windows of the measured spectra on each scan with those of a high-resolution solar spectrum, in order to find the optimal wavelength grid per window. A similar algorithm has been implemented for satellite data retrieval, for example (van Geffen and van Oss 2003).

Secondly, a flux calculation is directly proportional to plume speed and height and has a more complex dependency on plume direction. Scanning DOAS-type monitoring has the possibility of constraining these parameters from the measurements themselves, but only under specific conditions (Johansson et al. 2009). For the SO₂ measurements of the January 2010 Nyamulagira eruption, the following difficulties were found:

• Only the Rusayo station provided measurements during the study period. Therefore, it was not possible to find a geometrical solution to the plume height and direction by combining measurements from different stations.

• The Rusayo instrument is a conical single-beam scanner, which means that it does not measure plume speed by the correlation method described in Johansson et al. (2009).

• Risk and logistical criteria determined the selection of sites for the instruments in the Virunga.The setup favours detection of emissions from Nyiragongo, located 10 km NE of Rusayo station, while Nyamulagira is located 18.7 km N. Thus, the measured emissions might come from any or both of these two volcanoes, requiring careful analysis to identify the actual source.

To overcome the problems caused these particular settings, we used a combination of measurements and modelling. The Weather Research and Forecasting (WRF) model developed by the US National Center for Atmospheric Research is a mesoscale meteorological model. It uses global circulation model (GCM) datasets as an input to give the initial and boundary conditions for an atmospheric simulation run where many parameters, including wind, are calculated (Skamarock et al. 2005). This is done by first specifying a geographical domain for the simulation. The meteorological input data come from the National Oceanic and Atmospheric Administration FNL dataset, which provides data with 1° horizontal resolution at 26 vertical levels, every 6 h. Besides the meteorological parameters, GCM static data, such as topography and land use, are also supplied.

The simulation for Nyamulagira discussed here was made for 1 December 2009 to 31 January 2010 over a domain with a horizontal resolution of 9 km for the outer domain and 3 km for the nested inner domain. The time resolution of the output was chosen as 1 h. The static geodata had a resolution of 30 arc sec (∼900 m at ground level). From each scan intercepting the plume, the angle-weighted centre of mass is calculated with the NOVAC software (Galle et al. 2010). In turn, a multi-core WRF run provides the wind vectors at 26 height (pressure) levels from ground to about 20 km a.s.l., above the station, with a typical average vertical resolution of ∼717 ± 300 m. Due to atmospheric scattering effects, an acceptable distance of observation is assumed to be up to 25 km, i.e. comparable to the station-volcano distance. The maximum plume height is restricted to altitudes lower than 10 km above ground, as evidenced by visual accounts of the eruption. Given that the locations of the volcano summit and station as well as the angular range of scanning are known, this information is sufficient to iteratively look for the combination of wind direction and height values in the vertical profile that is compatible with the observed centre of mass of the plume, within a tolerance level of ±15° in plume direction.

The double-source/single-station configuration of this case demands cross-matching of the results from independent analyses for each volcano, using the method described here. The retrieved wind vectors are classified in four groups according to their inferred origin: (a) Nyamulagira, (b) Nyamulagira or Nyiragongo, (c) Nyiragongo and (d) ‘Undetermined’. For the latter case, a calculation of the wind parameters assuming a plume height at Nyamulagira’s summit level is performed. The distribution of these calculations is presented in Fig. 24.

Fig. 24

Distribution of the inferred origin of the plume according to the method described in “Appendix 5” (left); box-chart plots of their corresponding emission statistics (right)

Finally, the gas fluxes are calculated with the NOVAC program by incorporating the a posteriori wind information interpolated in time. A quality assessment of the individual scan results is done by estimating the percentage of the cross section of the plume captured within each scan.