Dr. Ben Holtzman, the Lamont Associate Research Professor at Lamont Doherty Earth Observatory, Columbia U., presents "Machine Listening reveals cyclic change in seismic source spectra in the Geysers geothermal field, California; can we learn to hear changes in the thermal-mechanical state?" at the MIT Earth Resources Laboratory.

"The earthquake rupture process reflects complex interactions of stress, fracture and frictional properties. In geothermal reservoirs, fluids, fluid pressure and thermal stresses add further complexity. Our unsupervised machine learning methods reveal patterns in temporal-spectral properties of seismic signals, based closely on those developed for music information retrieval, voice recognition and image analysis, using the spectrogram instead of the waveform. As an introduction, I will present results from human-listening experiments that are analogous to our unsupervised audio-based methods. Unsupervised learning involves identification of patterns based on differences among signals without any additional information provided to or training of the algorithm. Results from 46,000 earthquakes of 0.3