The research will focus on new detection and analysis of seismicity in subduction zones: Shallow seismicity in Colombia and Costa Rica. The detection will be performed with template matching techniques and attemped with convolutional neuronal networks (ConvNetQuake). The research will also involve relocation and magnitude estimation. The new catalogs will serve as a basis for b-value estimates and eathquake source study (stress drop and radiated energy) relating these metrics with the tectonic context of the subduction zones.
Now that most interns have gone back home after their summer internships I have all the audience for myself. It was not the original plan though. It’s been a long time since my last post, I first postponed my blog duties for a week and then another and so on, but I must fulfill my duty now.
As I have been gathering information and putting earthquake catalogs from my study area in Costa Rica together I convince myself more and more that this place is one of the best to study subduction processes in the world. The Cocos and Nazca plates subduct below the Caribbean plate at a rate of around 80 millimeters per year, producing megathrust events at a recurrence interval of about 50 years. You can see the study are in the map. This is the Frankenstein of maps by the way. It is a mixture of the Generic Mapping Tools, Python Basemap and my software best friend: Power Point. First, I plotted the topography ad bathymetry from the GTOPO30 database using GMT and adding illumination from the NW. Also in GMT, I created the scale bar and the compass Rose as well as the coastlines and country borders. The next step was to plot the locations of the seismological stations (20 land and 14 OBS). Furthermore, I added the Middle American Trench from the Active Tectonics of the Andes database by Veloza et al., 2012. Then I wanted to show where this region is in the world but adding a submap. Doing so in GMT seemed quite involved and the syntax didn’t help me a lot, so after a few tries I decided to do the small map in Basemap because I had done it in the past and yes, it was easy. But then I had to put the two maps together, to do it, I loaded them into a Power Point file just to notice that I had forgotten to add some labels to the countries, so I just added them in Power Point. Quite a process for this map.
On the other hand,I started building my codes for template matching. It gave me long days of frustration because of the long times of computing cross correlation and the amount of memory it requires. Therefore, I crashed the computer many times throughout the week. As I have been coding, I have reviewed many papers on template matching and other detection methods. I feel that when I was starting the summer I had a crude idea of how template matching works. By coding it myself my understanding of the method has increased a lot. Also, dealing with what I consider large amounts of data gives me a sense that my skills are really evolving.
Week three at Harvard goes with an elevator pitch to highlight the importance of increasing the number of earthquake detections!. I wanted to focus on the non-seismological community, so I wanted to avoid tthe technicalities, that is why I started with stunning numbers: Did you know that since the moment you started reading this post there have been around 35 magnitude 2 earthquakes around the world? Yes, but just about half of them are detected and actually go to earthquake catalogs. I want to convince people that we have to pay atttention to small earthquakes that we do not feel and that do not cause damage. The motivation to do so is that we can learn from the small ones how the big ones occur and to get prepared for those ones. It is common knowledge that nowadays more data is being recorded than can be processed, so I want to highlight the usefulness of applying effective, automatic computationally efficient methods to detect more of those earthquakes that we do not feel such as template matching and deep learning. I always feel comfortable talking about seismology regardless of the public, but in the case of this elevator pitch I want to be thought provoking for those who are not in earthquake science.
On the other hand, data processing has been slow but I feel galvanized by new ideas that have come up during the meetings with my mentor. We have been integrating all available data sets of previous studies in Costa Rica and we have noticed some interesting patterns that we would like to explore deeper.
PS: It happened to me that I had to rewrite this post because I was trying to include a photo, when I did, the photo fully replaced the text and deleted it. I found no way no recover it. This time Ctrl+Z didn´t save me so I had to rewrite the post. This isn't as good and as long as the one I lost, I apologize.
After the orientation week I had to go back to Colombia for a few weeks to go on a fieldtrip, that is the reason for my delayed start. Now, while I am sitting in the office let me tell you what my project is about. The goal is to take a look at data collected near the Nicoya peninsula in Costa Rica (a really well studied zone) by amphibious experiments during the early 2000's.The data is part of the Costa Rica Seismogenic Zone Experiment (CREISZE), a multi-institution scientific project, so there was a big deal of work made out of it. At first, I was overwhelmed by the amount of SAC files that I had to deal with. One issue was that when I arrived and the first couple of days my advisor was out of town and the database had not been properly transferred to my machine, but I wasn't aware of that, so I was fighting to make sense of the data. Once we clarified the issue, I ended up with 20 times more data than when I started but made a clear idea of the database organization. One of the advantages is that because these data is considered a bit old, nobody is loooking at it anymore, so my advisor and I feel more freedom about what we can do. The first step was to make a clear idea of the database, constructing plots of station and data availability. What we want to do now is to perform template matching detections based on catalogs of previously detected events and then attempt deep learning detection methods. Once we have a refined catalog we can move forward to estimate static and dynamic parameters of these earthquakes and relate them to the structure and dynamics of the shallow subduction zone offshore Costa Rica.
Sidenote: The Boston area is amazing, and very bike-friendly so I have enjoyed going around the city with my bike, which I brought from Colombia. At the beginnig I was afraid that the building in which I work was going to be empty because of summer vacations, but this very same building turns out to be the Harvard Museum of Natural History and there are always lots of people around and also the campus is full of tourists from all around the world.