资源环境科技发展态势分析平台

A method to locate tremor with an unclear onset is introduced. The method maps envelopes of cross-correlated records from pairs of seismographs to signal likelihoods using Bayes' theorem. The likelihood time series are then back projected to obtain likelihoods of source location in space. Assuming that information derived from different station pairs is independent, the joint likelihood for all station pairs is the product of all individual station pair likelihoods. Its peak and spread give the most probable source location and its uncertainty. Robustness of the method has been evaluated with synthetic tests. The method recovers true location within 0.5 km with realistic noise characteristics in synthetic data. Uncertainty estimates are consistent with location deviations for sources inside the seismic network. When applied to real data from Katla and Eyjafjallajokull volcanoes in southern Iceland, their likely tremor source is also recovered within 1 km.

Plain Language Summary Locating seismic sources is an important task in seismology. An accurate location of a volcanic event, for example, helps to understand processes within the volcano and sometimes to predict an eruption. Unlike earthquakes that usually manifest themselves as impulses in seismograms, many Earth processes, for example, continuous volcanic tremor, produce complex seismic signals without a clear onset. The timing of such processes cannot be measured. Therefore, traditional earthquake location methods fail. This article introduces a method to locate such processes. It maps the likelihood that a source occurs at any given location based on correlations of tremor recordings that are sensitive to the difference in propagation times of waves from the source to the two correlated recorders. The peak of this map is where the source location is most likely and its width provides information about its uncertainty. This is a major advance since most methods only yield an estimate of the location, not its uncertainty. The method has been tested with synthetic data that simulate realistic observational geometry and noise processes at volcanoes and is able to recover the true source location to within half a kilometer. Applications to real data at Katla and Eyjafjallajokull volcanoes in southern Iceland indicate similar accuracy.