Independent Component Analysis as a tool for analysing volcanic deformation


One of the challenges of analysing interferograms is distinguishing interesting, geophysical signals, such as earthquake or volcano deformation, from each other, and from atmospheric noise. Robust tests for the independence of geophysical signals are also important for establishing causal links between magmatic, hydrothermal and tectonic processes.

Independent Component analysis (ICA) is a computational signal processing method that decomposes a mixed signal into components that maximise signal independence in either space or time. This relies on the idea that as more and more independent signals are mixed together, their sum gets closer to a Gaussian distribution – so the parts of a mixed signal that are ‘interesting’ will be the least Gaussian. Finding these components is useful for analysing the relationship between the different deformation processes. Signals caused by the same process can be identified by searching for ‘clusters’ in the identified components. Real geophysical signals are very much more likely to cluster than atmospheric signals, so this is a useful approach both for distinguishing between true deformation and atmospheric noise and for analysing the relationships between geophysical signals.

I’m interested in the application of such blind source separation methods in volcanology, whether applied to InSAR, GPS, gas measurement etc., and would welcome any chances to collaborate.

Satellite data show alteration of cloud droplets downwind of degassing volcanoes in pristine oceanic regions

 COMET ‘highlight’ article April 2014 – March 2015

Satellite data show alteration of cloud droplets downwind of degassing volcanoes in pristine oceanic regions

Aerosols affect how reflective the Earth is by either absorbing and scattering solar radiation directly, or by modifying the properties of clouds. However, we are still uncertain of exactly how aerosols change cloud properties, how they affect global climate, and therefore how they impact on climate change.

It is more difficult to make satellite measurements of the effects of aerosols from passively degassing volcanoes deep in the lower atmosphere than from explosive eruptions that inject gas and aerosol up into the stratosphere. However, the impact of such ‘background’, volcanic activity is increasingly thought to be important to atmospheric processes. Prior to our study, measurements of a volcanic impact on cloud properties had been made only during a few episodes of elevated degassing.

We use data from three independent satellite sensors (MODIS, AATSR, CERES) to examine differences in cloud and aerosol properties upwind and downwind of isolated volcanic islands. By comparing this information with that from islands without active volcanoes, we can see how volcanic emissions are affecting the clouds.


Figure 1. Multiannual mean Aerosol Optical Depth from NASA’s MODIS Aqua Collection 6 dataset. Darker red indicates higher aerosol burden. The volcanoes and islands in our study are all in regions of low aerosol optical depth and may therefore be representative of the pre-industrial atmosphere.

By analysing a decade of satellite measurements of aerosol and cloud properties, we demonstrate that these volcanoes have a long-term net impact on cloud properties. Downwind of the volcanoes, the concentration of aerosol is higher and the cloud droplet size is lower than upwind. Top of atmosphere shortwave radiation flux is also higher downwind of the volcanoes, as smaller droplets tend to be more effective at reflecting solar radiation.

This was the case for a range of eruptive styles including high flux degassing (Kilauea), Strombolian eruptions (Yasur) and minor explosions (Piton de la Fournaise). Measurements of aerosol effects at isolated volcanic islands may now be the closest analogue to the pre-industrial atmosphere, and offer a rare chance to observe atmospheric processes as they would have been before the industrial revolution.


Figure 2. Aerosol optical depth (a & b) and cloud effective droplet radius (c & d) plots for Piton de la Fournaise between 2002 and 2008. Data are rotated according to wind direction at the height of emission, so that for each panel the upper quadrant is downwind of the volcano and the bottom panel is upwind (i.e. the arrow shows wind direction). Panels a & c show aerosol and cloud properties during quiescence, while b & d show elevated aerosol and smaller droplets during minor explosive activity.

You can see a short video describing the research on the NASA Goddard Space Flight Center YouTube channel.


Ebmeier, S. K., A. M. Sayer, R. G. Grainger, T. A. Mather, and E. Carboni (2014). Systematic satellite observations of the impact of aerosols from passive volcanic degassing on local cloud properties. Atmospheric Chemistry and Physics, 14, 10601-10618, doi:10.5194/acp-14-10601-2014.

Co-eruptive deformation at Calbuco, April 2015

On the 22nd April Calbuco volcano, Chile, erupted for the first time since 1972, with very little warning. Plumes of volcanic ash reached heights of 16 km on the 22nd and up to 17 km in a second, longer duration eruption that began in the early hours of 23rd April.  A third eruption occurred on the 30th April with a ash heavy plume that reached heights of 5 km and lahars on the slopes of Calbuco.   Large volumes of SO2 were released during the first days of the eruption


Sentinel-1 interferogram showing subsidence at Calbuco during eruptions on the 22-23rd April 2015. Each full colour cycle represents ~2.8 cm of motion away from the satellite.

Several thousand people were evacuated from villages closest to Calbuco .   Ash fell over an area extending from the west coast of Chile to the east coast of Argentina, and grounded air traffic in Chile, Uruguay and Argentina.

There appears to have been only a very brief period of precursory activity before the 22nd April eruption.  No deformation was measured in Sentinel interferograms from the month before the eruption up until the 21st April.  The first co-eruptive interferograms (14th – 26th April) at Calbuco show subsidence on the western flank of the volcano of ~ 12cm.  The topography of the volcano’s crater has also changed, possibly due to ice melt as well as the eruption.

For more information about satellite observations of Calbuco, see the COMET webpage, here.

Statistical link between deformation and eruption?

This is a reposting of an article from the STREVA website from last year

In a study published today,1 we demonstrate the statistical link between volcanic eruptions and measurements of deformation made from satellite radar.  For this, we used a global synthesis of 18 years of measurements to calculate diagnostic test statistics that strongly associated satellite measurements of deformation with volcanic eruptions: 46% of deforming volcanoes erupted, while only 6% of non-deforming volcanoes (‘false negatives’) did.

These results demonstrate the usefulness of satellite deformation measurement (InSAR) for volcano monitoring and its potential for hazard assessments, especially in inaccessible settings.  The launch of the European Space Agency’s first Sentinel satellite (expected in early April 2014) will increase the frequency of radar measurements, greatly improving the potential of InSAR measurements for volcano monitoring.   

Why do we need to know about volcano deformation?

Volcano deformation, measurable from space, along with seismic activity and gas emissions, is an important indicator of a volcano’s level of activity: volcanoes deform before, during and after most eruptions.

A broad range of processes can cause deformation at volcanoes, including magma movement, gas escape, cooling and crystallisation and the settling of fresh deposits.   Surface uplift is often interpreted as magma movement or increased pressure and is sometimes treated as an eruption precursor, although episodes of unrest do not necessarily end in eruption.

Traditional ground-based instruments for measuring deformation are only in operation at a small percentage of volcanoes worldwide, generally those that are near centres of population and have erupted recently.  Satellite measurements of deformation have massively increased the number of volcanoes where deformation measurements can be made.  At well-studied volcanoes, such as Etna or Kilauea, such measurements contribute to an understanding built up from a range of parameters over many decades.  In remote or inaccessible regions, however, satellite deformation measurements over the last few years are often the only observations available. Understanding what they might mean in terms of volcanic hazard is therefore very important.

Why do we use satellite data?

A lot of what we know about volcanoes comes from detailed studies of a few, very active volcanoes.  Furthermore, the time period over which we have been observing volcanoes captures only a very small fraction of their cycle of activity.  Our ideas about indicators of unrest, like gas emission or deformation, are therefore biased towards volcanoes with frequent historical eruptions and brief repose periods.  Because their coverage is global and observation repeat time is independent of levels of activity, satellite measurements have the potential to redress this bias.

However, for satellite data to be truly unbiased, measurements of volcano deformation must be made systematically, with both positive and `null’ results reported.  Although some recent work has considered uncertainties for individual volcanoes and lack of deformation, where it is measured,2 the vast majority of published satellite deformation studies report only measurements of deformation.  Overall, 59% of volcanoes covered by individual studies were shown to deform, relative to just 17% of those included in regional surveys.1

We therefore limited our synthesis to the 540 volcanoes covered by regional deformation studies.  Our diagnostic test statistics were calculated for the subset of volcanoes (198) where measurements have been made over the full 18 years that satellite instruments have been available.

Interpreting satellite measurements of deformation

To calculate generally applicable statistics for what is a highly complicated dataset, we have made our classifications as simple as possible.  For example, we classify each volcano in a systematically studied region as either ‘deforming’ or ‘not-deforming’ and either ‘erupting’ or ‘not erupting’ for a particular time period.  We find that volcanoes that erupted between 1992 and 2010 were 4 times more likely to deform than not (positive likelihood ratio.)

The reality of deformation and eruption is of course much more complicated.  Some volcanoes have multiple discrete episodes of deformation or eruption within a relatively brief time-period.  Deformation can be associated closely with eruption, such as by pressurization of magma, or with processes not necessarily driven by volcanic activity, such as gravity-driven spreading.  Satellite measurements may also miss highly localised, short-lived or low magnitude deformation.

Our classifications do not distinguish between deformation before, during or after eruption, and therefore do not imply a causal link.  In order to test the usefulness of deformation as a precursor to volcanic eruption, rather than the association between deformation and eruption presented here, we need frequent measurements during all stages of a volcano’s eruptive cycle.  New satellite instruments being launched in the coming months will improve temporal coverage to a level where this will be achievable for some volcanoes.

The causes and characteristics of deformation depend on tectonic settings, rock compositions and repose periods.  The interpretation of satellite observations of deformation (or lack of deformation) should, therefore, take these factors into account.  For example, a greater fraction of stratovolcanoes in arc settings erupt without satellite observations of deformation compared with shield volcanoes on rifts or hotspots.   This has implications for the usefulness of satellite measurements for hazard assessment in different contexts.

By using a near-global dataset, our research draws out both regional differences in volcano deformation and a general association between deformation measurement and eruption.


1.     ‘Global link between deformation and volcanic eruption quantified by satellite imagery’ by J. Biggs, S.K. Ebmeier, W.P, Aspinall, Z. Lu, M.E. Pritchard, R.S.J. Sparks, T.A. Mather in Nature Communications

2.     Ebmeier, SK, Biggs, J, Mather, T & Amelung, F 2013, ‘On the lack of InSAR observations of magmatic deformation at Central American volcanoes’Journal of Geophysical Research: Solid Earth, vol 118., pp. 2571-2585

Volcanic Impacts on clouds – NASA outreach video

This is NASA’s Goddard’s summary of our paper about the impact of passive volcanic degassing on cloud properties.  You can read more about it in the original paper, here.

Steady flank movement at Arenal, Costa Rica

Satellite measurements of deformation at Costa Rica’s most active volcano, Arenal, show that part of its western flank is moving downslope at a rate of about 7 centimetres every year. We think that this is caused by the weight of lava effused from the volcano over the last 40 years, and speculate that movement is along the plane between material laid down before and after Arenal’s reactivation in 1968. Monitoring the deformation of Arenal’s edifice is important both for understanding the way volcano stability responds to prolonged periods of eruption and for assessing the risk posed by the volcano to the surrounding population.

DSC01162Arenal has been erupting almost continuously since 1968, when it reawakened after centuries of dormancy. Three days of blast eruptions then covered about 200 square kilometres with ash and destroyed two nearby villages. Today about 7000 people live within approximately 6 kilometres of the volcano, mostly in the town of La Fortuna, where a significant proportion of the population are to some extent dependent on hot spring or volcano related tourism for their livelihood. Over the past 40 years Arenal has erupted almost continously, with the vast majority of new rock added to the western flank of the volcano, making it increasingly asymmetrical and increasing its total volume by about 4%.

Interferometric Synthetic Aperture Radar (InSAR) is a satellite-based measurement technique that allows us to measure the movement of the earth’s surface to centimetric precision, or better. Interferograms are constructed from radar images acquired on different dates and map change in phase over time. After correction, this difference in phase can be used to determine the movement of the earth in the satellite’s line of sight.


Figure 1: a) Map showing the location of Arenal on the Central American Volcanic Arc. b) the average deformation rate measured at Arenal between 2005 and 2009 using RadarSat data. Each complete transistion in colour scale represents 2.6 cm/year of motion.

Over 40 interferograms constructed from two different wavelengths of satellite data were used to measure deformation at Arenal between 2005 and 2009. One of the strengths of InSAR as a technique is its high spatial resolution, which allows us to develop a detailed picture of the shape of the deformation signal (Figure 1). We gained more information about the actual direction of slope movement, as opposed to that in the satellite line of sight, by using two different satellite look directions to resolve the motion into vertical and East-West components (Figure 2). We used this evidence about direction of deformation, along with analysis of the signal shape and variation through time to consider several possible causes for the deformation signal, and concluded that the most likely mechanism is gravity-driven slip.


Figure 2: Profiles though Arenal running East-West, showing a) resolved angle of deformation as arrows with length proportional to magnitude of deformation and b) estimated boundary between pre- and post 1968 eruption deposits.

  The rate of deformation at Arenal is unusually high compared to other gravity-driven volcano deformation described in the literature, especially for such a young volcano. This suggests that the plane of failure at Arenal is already well established. Our data suggest that for the period of measurement (2005-2009), slip is constant. However, it is as yet unclear whether this deformation is stabilising or destabilising the slope. Events such as the intrusion of magma into its edifice or a large earthquake could be expected to have a destabilising effect on Arenal’s western slopes and may trigger rapid flank collapse. Detecting any change in the rate of motion on Arenal’s western flanks is therefore important for assessing hazard posed by the volcano.