Earthquake data collected from the Geysers geothermal field from the period 2006 to 2010 was studied. An artificial neural network (ANN) based autopicker was developed to study and compare the picks with available autopickers in use at Lawrence Berkeley National Lab (LBNL) as well as in house autopicker at USC. The results indicate the following: 1. The ANN autopicker is able to generate good picks in most situations where other autopickers work. 2. The ANN autopicker is able to pick in situations where the other autopickers fail to generate good picks or do not generate any picks at all due to noise. 3. STA/LTA ratio and frequency based attributes show highest weight in the trained ANN. Despite the low misclassification error observed in the final results (3.7%), the ANN autopicker failed in some situations. The next step is to incorporate improved attribute sets and look at a hybrid neuro-fuzzy approach to tide over the limitations of the present network.

This content is only available via PDF.
You can access this article if you purchase or spend a download.