Volcanic earthquakes: their use for eruption forecast, detection, and classification

Abstract

Volcanoes pose dangerous threats, so their activities must be monitored. One of the methods to monitor their activities is by monitoring earthquakes or seismic events generated by volcanic or magmatic activities. In terms of seismicity, volcanoes exhibit various seismicity patterns that reflect the current states of the volcanoes. Generally, there are three standard practices conducted by volcano observatories to monitor volcano seismicity. The three standard practices are (1) monitoring the change in seismic event rates or seismic energy rates, (2) monitoring hypocentre migrations, and (3) observing the earthquake types occurring in specific periods. This thesis studies those standard practices and proposes new methods or procedures that are intended to improve volcano monitoring efforts.

The first proposed method is forecasting the onsets of volcanic eruptions using migrating seismicity, which is related to the second standard practice. The exponential or hyperbolic trends of the seismic amplitude time series during the migrating phase can be fitted with the Failure Forecast Method to forecast the onsets of volcanic eruptions retrospectively. This method is applied to synthetic data and field data of the January and October 2010 Piton de la Fournaise eruptions. The results show that the onsets of the eruptions are contained within the forecasted eruption time windows.

The second proposed method is related to the first standard practice, earthquake detection. This method is based on the STA/LTA algorithm with minor modifications. Using the modified method, more representative waveforms that include the arrival and coda waves can be obtained. The second proposed method is applied to data sets from Merapi volcano, producing a seismic catalogue that is used later as the inputs for the third and fourth proposed methods/procedures.

The third and fourth procedures are related to the third standard practice, which is earthquake classification. First, the k-means algorithm is used to find natural groupings or clusters of a set of earthquake waveforms that are obtained by applying the second proposed procedure. Later, the resulting clusters are used as ground truths for classification. The classification is done by building an Artificial Neural Network (ANN) model. The ANN model is trained and validated using the ground truths obtained from clustering steps. The results show that the validation accuracy of the ANN model is 96% – 98%.

The proposed procedures for detection, clustering, and classification are finally tested using new (unseen) data. The data for the final test are seismic time series recorded at Merapi volcano from December 27, 2020, to January 7, 2021. The results show that there is a positive temporal correlation between the observatory’s number of detections and the number of detections obtained by the proposed procedure. In addition, the accuracy of the ANN model is 95%, indicating that there is no overfitting problem.

Despite some limitations, the new procedures proposed in this thesis may improve the existing standard procedures used by volcano observatories. These improvements can help volcano observatories make timely and proper decisions during volcanic unrest.