Visualizing Earthquakes

These data consist of the dates, times, locations, depths, and magnitudes of all 3+ magnitude earthquakes in California, Nevada, and Oregon for approximately the last 30 years. (I'm going to have to dig up my online source again - it's definitely possible to get up-to-the-minute data. It's also possible to set a lower threshold for the magnitudes to include. (DTL - Possibly ) I used this example in Stat 133 to demonstrate how we can create KML animations in Google Earth; see the attached assignment and screenshot. For this assignment there was no specific scientific question, just to describe the characteristics of the process that were apparent visually. (For example, you can clearly see fault lines in the animation, and it's also interesting to see how incredibly common the low-magnitude earthquakes are.) There are a number of ways to spin off from this, for example to take the qualitative observations from the animation and think about how they might inform modelling choices for fitting something like a self-exciting point process model. I think there are also many simpler explorations/discussions that could be pursued, such as
  • how to scale the size of the points in KML to represent the magnitude of the earthquakes - this raises some questions about human perception and how to honestly represent the data
  • discussing extreme value distributions and estimating parameters related to the occurrence of various extreme events
  • discussing censored data
  • ignoring the magnitudes of the events and fitting a surface representing the conditional intensity -- where are the most earthquake-prone locations?
  • looking for evidence of triggering of aftershocks
  • looking at the distribution of waiting times between earthquakes within a certain radius of a particular location. I used this as part of an example discussing sensitivity analysis when I was teaching Bayesian statistics. We went through the process of eliciting a "class prior" for the mean waiting time between events near Berkeley and compared the results to using a noninformative prior -- see SensitivityAnalysis.R.