Grid-search method for STA/LTA parameters tuning: an application to Stromboli Explosion Quakes

The collection of a significant catalog of seismo-volcanic data involves the selection of relevant parts of raw signals, that can be automatized by using the Short-term over Long-term Average (STA/LTA) method. Since it is parametric, the common approach to the choice is the adoption of literature-suggested parameters. To overcome these limitations, we propose a methodology for the automatic selection of STA/LTA parameters able to optimize the extraction of local events from a seismo-volcanic raw signal. The parameters are found by a grid search over an index named Quality-Numerosity Index (QNI) that measures the accordance in the automatic cuts and the consequent quantity of triggered seismo-volcanic events with the ones suggested by a human expert. The method was applied in the volcano domain, for the specific application of Explosion Quake signals extraction in Stromboli Volcano. Experiments have been conducted selecting a subset of the dataset as training where to search for the best parameters, which were subsequently adopted in a test set. The results demonstrate that the selected parameters significantly improve the quality of the extraction when compared to those extracted by adopting the parameters indicated in the literature.