This had to be the fastest summer of my entire life. Where did these four months go??! I can't believe that this is my last week here in Chapel Hill, and that I'm flying straight back to school on Saturday morning. Time flies when your days are full, but I wouldn't have it any other way. My internship this summer challenged me in ways I hadn't been challenged before, and I am so happy that I was chosen to join the IRIS community.
Before I leave, though, there's still plenty more to do. Last week I focused on running a cross-correlations on my data in order to create groups of similar waveforms and perform a cluster analysis for each year. Jonathan gave me a bunch of code to figure out, which took me a while to get through. After playing around with it for a bit, I was able to use it to read in all my events, remove any that may be repeats not caught through the previous round of sorting, and compare the shape of the acoustic wave with all others that year. The code produces a correlation coefficient, which is then used to create clusters of events that are statistically similar. In R, I'm using a function called hclust(), which stands for hierarchichal cluster analysis. I spent quite a while researching this topic, both on the internet and in some statistics and data analysis textbooks, and the methods behind it are pretty confusing and complex. Basically, it just means that I'm making groups--however many I choose--of my data based on which signals look similar. This uses a top-down approach, where you start with one large group of waveforms and divide it into clusters that are more or less like each other. Each cluster is further divided until you've reached the point where all the data is grouped, which can be illustrated with a large dendrogram plot like this:
These are all the good events saved from 2010 (the sparsest year--easiest to see what's going on) grouped through hierarchical cluster analysis. The "Height" label on the y-axis is the correlation coefficient, so the groups that are higher up on the plot are more alike. From here, I split all of these into 6 different groups:
Now it's clear to see that each individual event fits into a group that is statiscally "alike." I've been going through each of the grouped events to see what they have in common and whether or not I can make any connections between the groups (which are sorted based on their waveforms--explosions will have different waveforms than thunder or mineblasts, for example) and where they're located on the stereonet (which is based on my original function to calculate arrival incidence vectors).
This week I'm continuing what I was doing last week...diving through my data looking for correlations or patterns or really anything that can help me better understand and predict the sources of these infrasound events. I've written codes to analyze the data in all different ways, comparing stereonet positions with cross correlation data, going back and looking at the original waveforms I made my picks on, extracting the dates and times from each of the events to see whether they occurred on weekdays or weekends, during working hours or late at night, etc. Mostly, I'd like to know more about the signals that have weird inclinations, or come from an azimuth angle for which there is no predicted/obvious source. With Danny's help, I'm trying to get a hold of someone who can provide a record of all mine blasts in New Mexico during these years--maybe a lot of these are coming from far-off mines. I've taken extensive notes as I've been going through my data, and am sad to say I haven't found anything extraordinary yet. And maybe I won't--it's something Jonathan and I have been talking about a lot recently, that maybe, after all of this, I won't be able to produce some clear, groundbreaking conclusion accurately solving for all infrasound sources in the area. There are so many factors that could affect the signals before they reach the array, so I can't expect to have all the answers, no matter how long I spend analyzing the data. It's a little frustrating, but I plan on moving forward with it and putting together the most comprehensive poster I can.
I will definitely have more work to do once I get back to Yale--I have to make the poster, finalize my conclusions, and I would like to turn all of this work into a paper, which is going to be a serious challenge seeing as how I'll also be working on my senior thesis this year. However, I love that I had a chance to experience an entirely new field of geophysics research this summer, and I feel like, for the first time, this project is truly mine. Sure, it's a little rough right now, and Jonathan could have probably produced the same results I achieved this summer in just a few weeks on his own, but I am proud of it. It has been invaluable participating in the research process firsthand from start to finish. I'm still just as (if not more) confused on whether or not I will pursue a Ph.D. program, but I do feel like I have learned an enormous amount about what it's like to be a research scientist.
You must be logged in to post a comment.