Volcanoes

A challenge in the analysis of multi-temporal Interferometric Synthetic Aperture Radar (InSAR) data is distinguishing and separating volcanic, tectonic and anthropogenic displacements from each other and from atmospheric or orbital noise.

Independent Component Analysis (ICA) is a computational signal processing method that aims to describe random variables as a linear combination of statistically independent components. Mixed signals are decomposed using the assumption that each constituent component has a non-Gaussian probability distribution. ICA is already widely used in the fields of medical imaging and has been applied to remote sensing applications including hyperspectral unmixing, cloud identification and detection of thermal hotspots.   

ICA has potential as a tool for exploratory analysis of InSAR datasets and in particular for assessing the relationships between deformation signals.  Deformation sources that do not share a causal mechanism are likely to result in independent displacement patterns, and as such will be decomposed into separate ICs.  Exploratory analysis requires a reliable method for assessing the statistical significance of the ICs.  I achieve this by dividing InSAR datasets into independent groups and using a cluster analysis of the spatial patterns identified as independent components.  This is likely to be particularly useful for interrogating the large volumes of satellite radar imagery, such as the Sentinel-1 archive, now available for monitoring geophysical signals. 

I present tests of the applicability of ICA to InSAR using synthetic data and application to Sentinel-1A archive images from examples of volcano deformation in Latin America.  These include co-eruptive subsidence associated with the April 2015 eruption of Calbuco (Chile), which was identified in spatial patterns found by maximising both spatial and temporal independence. In contrast, spatial patterns and rates of lava subsidence were retrieved using ICA analysis of interferograms from Parícutin lava fields (Mexico), and found to be consistent with previous observations. 

These prototype examples demonstrate that the combination of ICA and cluster analysis is useful (1) identifying geophysical signals caused by tectonic, volcanic or anthropogenic processes and (2) testing the independence of geophysical signals. 

Remote sensing observations (including InSAR) have proven their worth in volcano monitoring and volcano science, for example expanding the number of known deforming volcanoes from 44 to over 230 in the last 20 years. While our sampling of global volcano deformation is still not complete, some characteristics of the duration, frequency, and magnitude of these events and their relationship to eruption are starting to emerge through the creation of databases of over 490 volcano deformation episodes at COMET and the Smithsonian Institution. Volcano deformation events can last for seconds to centuries and span from a few meters to hundreds of km, and thus can be challenging for any single satellite to observe. To address this challenge, the 2012 "Santorini report" from the "International Forum on Satellite EO and Geohazards" suggested an integrated, international, global remote sensing geohazards monitoring effort for disaster risk management (Bally et al., 2012) that could leverage the constellation of more than a dozen satellites that can observe volcanoes. We are working with others, including the Committee on Earth Observing Satellites (CEOS) to realize this vision by developing a strategy, using the lessons learned thus far about volcano deformation events from a variety of SAR satellites, to target volcanoes depending on their type of activity, geographical region, and environmental setting. For example, volcanoes thought to be dormant may only need to be observed a few times per year, while very active ones should be viewed as frequently as possible. Volcanoes in densely vegetated areas require different observing modes (e.g., long wavelength radars or frequent observations of short wavelength radars) than those in more arid areas. Some volcanoes have activity concentrated in small areas (e.g., inside a crater) and require higher spatial resolution (better than 1 m/pixel) than some large caldera systems that require observations spanning more than 100 km. For example, we find that while high resolution Spotlight mode SAR data reveals new deformation in some areas that cannot be detected with lower resolution sensors (like Colima volcano, Mexico), in other areas (like Villarrica volcano, Chile), the small spatial footprint of the SAR images misses the broader deformation pattern. Sentinel-1 mission represents a major step forward in terms of volcano observation, and is yielding new results thanks to its frequent repeat and dedicated mission. We describe areas where existing Sentinel-1 observations are proving useful, even in otherwise challenging areas (e.g., the tropical volcanoes of Colombia) while in other places, a different observing strategy might be needed. For example, VV polarization data acquired every 24 days is often incoherent over the most dangerous volcanoes in southern Chile while shorter time period observations or using HH polarization provide better data quality. We report on our efforts to optimize satellite observations from the international constellation of satellites to provide the best observations of deformation for each of the world's ~1400 subaerial volcanoes that have been active in the last 10,000 years.

Nevado del Ruiz (NRV) is part of a large volcano complex in the northern Andes of Colombia with a large glacier that erupted in 1985, generating a lahar killing over 23,000 people in the city of Armero and 2,000 people in the town of Chinchina. NRV is the most active volcano in Colombia and since 2012 has generated small eruptions, with no casualties, and constant gas and ash emissions. Interferometric synthetic aperture radar (InSAR) observations from ascending and descending track RADARSAT-2 data show a large (>20 km) wide inflation pattern apparently starting in late 2011 to early 2012 and continuing through at least 2015 at a LOS rate of over 3-4 cm/yr. Volcano pressure volume models for both a point source (Mogi) and a spheroidal (Yang) source find solutions over 14 km beneath the surface, or 10 km below sea level, and centered 10 km to the SW of Nevado del Ruiz volcano. The spheroidal source has a roughly horizontal long axis oriented parallel to the Santa Isabel – Nevado del Ruiz volcanic line and perpendicular to the ambient compressive stress direction. Its solution provides a statistically significant improvement in fit compared to the point source, though consideration of spatially correlated noise sources may diminish this significance. Stress change computations do not favor one model over the other but show that propagating dikes would become trapped in sills, leading to a more complex pathway to the surface and possibly explaining the significant lateral distance between the modeled sources and Nevado del Ruiz volcano. Since September 2015 COSMO-SkyMed (CSK) and TerraSAR-X (TSX) data track the dome growth in the summit crater, which appears to have occurred in at least two episodes, the first one ended in early 2016 and the second one is ongoing since August 2016. InSAR time series may indicate deflation coincident with dome extrusion episodes, though deflation deformation amplitudes are low. We compare the InSAR (including Sentinel-1) and dome growth analysis (CSK, TSX, and possibly TDX) with contemporaneous in-situ volcanic activity observations to guide future volcano system analysis of Nevado del Ruiz.

Located 48 km offshore, White Island is New Zealand's most active volcano with eruptive activity for the past 150,000 years. Despite ~70% of the volcano being beneath the sea, the main crater is host to a vigorous hydrothermal system and acid crater lake. The island was the scene of one of New Zealand's major volcanic disasters in 1914 when the collapse of the south west corner of the crater wall caused a debris avalanche killing 11 miners. Between 1967 and 2009, 9 long-term deformation events were detected from levelling data within the crater floor. Accompanying these events were a range of activity from small eruptions to more passive degassing. The most recent eruption, on April 27^th, removed ~15 m of lake floor sediments and much of the crater lake. Here we use high resolution TerraSAR/TanDEM-X SAR data acquired since 2015 to track changes in the Lake level in the build up to the eruption and to generate a deformation time series. Preceding the eruption, we observe uplift of the crater floor, consistent with the pressurization of the hydrothermal system, and at the same time evidence for continued creep of the south west crater wall which collapsed in 1914. In the months running up to the eruption both the uplift and movement of the crater wall decrease and, at the same time, we see a decrease in the lake level. The eruption deposited a layer of ash across the crater floor which we assess using a combination of field measurements, coherence loss and changes in the radar amplitude. Following the eruption, as the lake began to refill, we see renewed motion of the crater wall accompanied localised subsidence around the crater lake. We suggest that changes to the hydrothermal system and crater lake are sufficient to alter the pore pressure triggering renewed motion of the crater wall.

Volcanic eruptions are among the most significant hazards to human society, capable of triggering natural disasters on regional to global scales. In the last decade, remote sensing has become established in operational volcano monitoring. Centers like the Alaska Volcano Observatory rely heavily on remote sensing data from optical and thermal sensors to provide time-critical hazard information. Despite this high use of remote sensing data, the presence of clouds and a dependence on solar illumination often limit their impact on decision making.

Synthetic Aperture Radar (SAR) systems are widely considered superior to optical sensors in operational monitoring situations, due to their weather and illumination independence and their capacity to detect cm-scale surface motion through interferometric SAR (InSAR) processing. Still, the contribution of SAR to operational volcano monitoring has been limited in the past due to the typically high data costs, long processing times, and low temporal sampling rates that used to plague most SAR systems. 

In this paper, we introduce the SAR processing system SARVIEWS, which takes advantage of recent improvements to sensor and processing technologies to automatically generate and seamlessly integrate SAR-derived hazard information into operational volcano monitoring systems. SARVIEWS utilizes ESA’s Sentinel-1 mission, providing free access to regularly observed SAR data over most volcanically active regions.

We will introduce the SARVIEWS database interface that facilitates the integration of the SARVIEWS system with the Sentinel-1 SAR holdings at the Alaska Satellite Facility. We will also present the core processing techniques within SARVIEWS that were designed to automatically generate a collection of SAR-based hazard products from Sentinel-1 and other available SAR data. Currently, this suite of products includes time series of geocoded and radiometrically terrain corrected images, change detection maps, differential interferograms with various temporal baselines, as well as associated coherence maps. The production of deformation time series is also planned and is currently in preparation. Finally, we will show how SAR-based hazard information is integrated into existing multi-sensor decision support tools to enable joint hazard analysis with data from optical and thermal sensors. Specifically, we will show an integration into the hazard analysis tools of V-ADAPT, Inc., a provider of decision support information for volcanic hazards (http://www.vadapt.net/).  

The presentation will showcase the performance of the SARVIEWS processing system using a set of recent natural disasters (both earthquakes and volcanic eruptions). We will also show the benefit of integrating SAR with data from other sensors to support volcano monitoring. Here, we will use examples from recent volcanic eruptions in Alaska and South America.