Andrew F. Bell

Precursors to dyke-fed eruptions at basaltic volcanoes: insights from patterns of volcano-tectonic seismicity at Kilauea volcano, Hawaii

To investigate the physical controls on volcano-tectonic (VT) precursors to eruptions and intrusions at basaltic volcanoes, we have analyzed the spatial and temporal patterns of VT earthquakes associated with 34 eruptions and 23 dyke intrusions that occurred between 1960 and 1983 at Kilauea, in Hawaii. Eighteen of the 57 magmatic events were preceded by an acceleration of the mean rate of VT earthquakes located close to the main shallow magma reservoir. Using a maximum-likelihood technique and the Bayesian Information Criterion for model preference, we demonstrate that an exponential acceleration is preferred over a power-law acceleration for all sequences. These sequences evolve over time-scales of weeks to months and are consistent with theoretical models for the approach to volcanic eruptions based on the growth of a population of fractures in response to an excess magma pressure. Among the remaining 40 magmatic events, we found a significant correlation between swarms of VT earthquakes located in the mobile south-flank of Kilauea and eruptions and intrusions. The behaviour of these swarms suggests that at least some of the magmatic events are triggered by transient episodes of elevated rates of aseismic flank movement, which could explain why many eruptions and intrusions are not preceded by longer-term precursory signals. In none of the 57 cases could a precursory sequence be used to distinguish between the approach to an eruption or an intrusion, so that, even when a precursory sequence is recognized, there remains an empirical chance of about 40% (24 intrusions from 57 magmatic events) of issuing a false alarm for an imminent eruption.

Statistical evaluation of characteristic earthquakes in the frequency-magnitude distributions of Sumatra and other subduction zone regions

[1] If subduction zone earthquakes conform to a characteristic model, in which persistent segments fail at predictable stress levels due to the steady accumulation of tectonic loading, historical seismicity may constrain the occurrence of future events. We test this model for earthquakes on the Sumatra-Andaman megathrust and other subduction zones using frequency-magnitude distributions. Using simulations, we show that Poisson confidence intervals correctly account for the counting errors of histogram data. These confidence intervals demonstrate that we cannot reject the Gutenberg-Richter distribution in favor of a characteristic model in any of the real catalogues tested. A visual bias in power-law count data at high magnitudes, combined with a sample bias for large earthquakes, is sufficient to explain candidate characteristic events. This result implies that historical earthquakes are likely poor models for future events and that Monte Carlo simulations will provide a better assessment of earthquake and associated hazards.

Heterogeneity: The key to failure forecasting.

Elastic waves are generated when brittle materials are subjected to increasing strain. Their number and energy increase non-linearly, ending in a system-sized catastrophic failure event. Accelerating rates of geophysical signals (e.g., seismicity and deformation) preceding large-scale dynamic failure can serve as proxies for damage accumulation in the Failure Forecast Method (FFM). Here we test the hypothesis that the style and mechanisms of deformation, and the accuracy of the FFM, are both tightly controlled by the degree of microstructural heterogeneity of the material under stress. We generate a suite of synthetic samples with variable heterogeneity, controlled by the gas volume fraction. We experimentally demonstrate that the accuracy of failure prediction increases drastically with the degree of material heterogeneity. These results have significant implications in a broad range of material-based disciplines for which failure forecasting is of central importance. In particular, the FFM has been used with only variable success to forecast failure scenarios both in the field (volcanic eruptions and landslides) and in the laboratory (rock and magma failure). Our results show that this variability may be explained, and the reliability and accuracy of forecast quantified significantly improved, by accounting for material heterogeneity as a first-order control on forecasting power.

The dilatancy–diffusion hypothesis and earthquake predictability

Abstract The dilatancy–diffusion hypothesis was one of the first attempts to predict the form of potential geophysical signals that may precede earthquakes, and hence provide a possible physical basis for earthquake prediction. The basic hypothesis has stood up well in the laboratory, where catastrophic failure of intact rocks has been observed to be associated with geophysical signals associated both with dilatancy and pore pressure changes. In contrast, the precursors invoked to determine the predicted earthquake time and event magnitude have not stood up to independent scrutiny. There are several reasons for the lack of simple scaling between the laboratory and the field scales, but key differences are those of scale in time and space and in material boundary conditions, coupled with the sheer complexity and non-linearity of the processes involved. ‘Upscaling’ is recognized as a difficult task in multi-scale complex systems generally and in oil and gas reservoir engineering specifically. It may however provide a clue as to why simple local laws for dilatancy and diffusion do not scale simply to bulk properties at a greater scale, even when the fracture system that controls the mechanical and hydraulic properties of the reservoir rock is itself scale-invariant.

Mode switching in volcanic seismicity: El Hierro 2011–2013

The Gutenberg-Richter b value is commonly used in volcanic eruption forecasting to infer material or mechanical properties from earthquake distributions. Such studies typically analyze discrete time windows or phases, but the choice of such windows is subjective and can introduce significant bias. Here we minimize this sample bias by iteratively sampling catalogs with randomly chosen windows and then stack the resulting probability density functions for the estimated b˜ value to determine a net probability density function. We examine data from the El Hierro seismic catalog during a period of unrest in 2011–2013 and demonstrate clear multimodal behavior. Individual modes are relatively stable in time, but the most probable b˜ value intermittently switches between modes, one of which is similar to that of tectonic seismicity. Multimodality is primarily associated with intermittent activation and cessation of activity in different parts of the volcanic system rather than with respect to any systematic inferred underlying process.

eScience Gateway Stimulating Collaboration in Rock Physics and Volcanology

Earth scientist observe many facets of the planets crust and integrate their resulting data to better understand the processes at work. We report on a new data-intensive science gateway designed to bring rock physicists and volcanologists into a collaborative framework that enables them to accelerate their research and integrate well with other Earth scientists. The science gateway supports three major functions: 1) sharing data from laboratories and observatories, experimental facilities and computational model runs, 2) sharing computational models and methods for analysing experimental and observational data, and 3) supporting recurrent tasks, such as data collection and running application in real time. Our prototype gateway has worked with two exemplar projects giving experience of data gathering, model sharing and data analysis. The geoscientists found that the gateway accelerated their work, triggered new practices and provided a good platform for long-term collaboration.

Volcanic eruption forecasts from accelerating rates of drumbeat long-period earthquakes

Accelerating rates of quasiperiodic “drumbeat” long-period earthquakes (LPs) are commonly reported before eruptions at andesite and dacite volcanoes, and promise insights into the nature of fundamental preeruptive processes and improved eruption forecasts. Here we apply a new Bayesian Markov chain Monte Carlo gamma point process methodology to investigate an exceptionally well-developed sequence of drumbeat LPs preceding a recent large vulcanian explosion at Tungurahua volcano, Ecuador. For more than 24 hr, LP rates increased according to the inverse power law trend predicted by material failure theory, and with a retrospectively forecast failure time that agrees with the eruption onset within error. LPs resulted from repeated activation of a single characteristic source driven by accelerating loading, rather than a distributed failure process, showing that similar precursory trends can emerge from quite different underlying physics. Nevertheless, such sequences have clear potential for improving forecasts of eruptions at Tungurahua and analogous volcanoes.

Crackling Noise in Digital and Real Rocks–Implications for Forecasting Catastrophic Failure in Porous Granular Media

‘Crackling noise’ occurs in a wide variety of systems that respond to steady-state external forcing in an intermittent way, leading to sudden bursts of energy release similar to those heard when crumpling a piece of paper or listening to a fire. In rock physics sudden changes in internal stress associated with microscopically-brittle rupture events lead to acoustic emissions that can be recorded on the sample boundary, and used to infer the state of internal damage. Crackling noise is inherently stochastic, but the population of events often exhibits remarkably robust scaling properties, in terms of the source area, duration, energy, and in the waiting time between events. Here we describe how these scaling properties emerge and evolve spontaneously in a fully-dynamic discrete element model of sedimentary rocks subject to uniaxial compression applied at a constant strain rate. The discrete elements have structural disorder similar to that of a real rock, and this is the only source of heterogeneity. Despite the stationary strain rate applied and the lack of any time-dependent weakening processes, the results are all characterized by emergent power law distributions over a broad range of scales, in agreement with experimental observation. As deformation evolves, the scaling exponents change systematically in a way that is similar to the evolution of damage in experiments on real sedimentary rocks . The potential for real-time forecasting of catastrophic failure obeying such scaling rules is then examined by using synthetic and real data from laboratory tests and prior to volcanic eruptions. The combination of non-linearity in the constitutive rules and an irreducible stochastic component governed by the material heterogeneity and finite sampling of AE data leads to significant variations in the precision and accuracy of the forecast failure time. This leads to significant proportion of ‘false alarms’ (forecast too early) and ‘missed events’ (forecast too late), as well as an over-optimistic assessments of forecasting power and quality when the failure time is known (the ‘benefit of hindsight’). The evolution becomes progressively more complex, and the forecasting power diminishes, in going from ideal synthetics to controlled laboratory tests to open natural systems at larger scales in space and time.