Deep learning for information extraction from earthquakes seismological data
Suggested elective course(s)
- Big Geodata Processing
- Advanced Image Analysis
Deep learning is an evolving scientific field that has proven superiority in big data applications in many fields. On the other hand, seismological data have been growing exponentially in the last decade, with more than 100 TB of data added yearly to the global seismic records. Analyzing such an amount of data is a challenging and time-consuming task for individuals and even institutes. In this topic, deep learning techniques will be used to automatically extract information about earthquakes from continuous seismological records. Recently developed deep learning algorithms will be used and tested on the seismological data from the Netherlands for information extraction and subsurface imaging. The earthquake information and possible subsurface models driven using deep learning approaches will be quantitatively compared with existing information and models to reflect on the feasibility of the methods and the suitability for robust automatic information extraction and the associated uncertainties.
Additional remarks
This topic requires strong programming skills and familiarity with deep learning.
In this topic, recently developed algorithms using TensorFlow, PyTorch, or other packages will be used to extract information about earthquakes. The target problem can vary from identifying earthquakes, “noise”, signal classification, phase extraction, source characteristics, or other properties of the seismological signals. A quantitative comparison will be considered to understand the limitations and the associated uncertainties in comparison with traditional existing information extraction techniques from the seismological data.