We will be using a variety of detection methods to expand the earthquake catalog and gain a better understanding of seismic evolution. We want to examine how this can relate to fault zone structures and processes. I will mainly be focused on using the Matched Filter Technique. Essentially, smaller earthquakes, such as foreshocks and aftershocks, can be vital to understanding the seismic evolution of a region. However, these events are so small that they are often simply not detectable by the human eye. This is where the Matched Filter Technique is applied. It works by starting out with a template. A template is a known earthquake in a particular region that has had its phases picked. This template is then compared to the continuous waveform data, and cross correlation between signals is computed. A cross correlation value ranges from -1 to 1. A value of 1 means that the events are a perfect match, and a cross correlation of -1 means that the events are exactly opposite in polarity. The events are then stacked. Using this method, we hope to provide a more complete catalog of earthquakes. For my project, my study region is the Sichuan region in China. I will be examining the Wenchuan earthquake that occurred in 2008. I hope to investigate whether or not there is evidence that this event caused remote triggering in several different regions throughout China.
These past couple of weeks have mostly been me going through more and more data using the Matched Filter Technique (MFT). I do not want to jinx myself, but I would like to think that I have finally figured out most of the nuances of the processes and my multitude of codes. To help myself, I have created a self-help document that is a step-by-step guide that details what works, the commands I need, what files I need and what did not work in the past. Creating this document has honestly helped a tremendous amount. It saves me the trouble of trying to struggle through the confusing lines of code each time that I try to run it.
Learning the nuances of microearthquake detection has been a long process. What confused me before and took me days to complete now takes me an hour to do. Whenever there is a bug in my code, I have realized that the problem is often times subtle and minor. This is good, because that means that my code is working and does not need major adjustments. On the other hand, this is also frustrating because the problem is often not immediately recognizable and may take a long time to recognize. For example, when I pick p-wave and s-wave arrivals the program called SAC saves the picks as “a” and “t0” respectively. However, my code refers to the p-waves arrivals as “t1” and the s-wave arrivals as “t2.” I did not realize this at first, and my code would not run because even though I had spent a decent amount of time identifying the arrivals, they were not labelled properly. I then realized that I had to change the name of the values in each file’s header. After this change, my code began to work.
Another frustrating issue that I ran across was related to self-detection. At the simplest level, what my code does is take a template and compare it to continuous data to see how similar it is to the template. It does this through computing the cross-correlation. Self-detection arises when the template recognizes itself in the continuous data. Therefore, wherever the template lies in the continuous data, a cross-correlation value of 1 is seen. Bewilderingly, when I first ran my code, I would get no self-detection. This was a red flag that something was not running correctly and was frustrating to see that after hours of running a code, it did not work properly. After a very long day, the issue was found: the time window for the data on the day of the template was incomplete. When editing my data, I had accidentally left a few stations in the continuous data that had very different time windows. These data had windows that were half the time of the rest of the data, and the template arrivals did not happen to be within this time window. My code, when looking at continuous data for each day, will only run on the smallest common time window. Hence, the code was not even looking at the time window in which the template and a lot of data were located. Once I fixed this issue, I finally started to get self-detections.
While I am still running the MFT, I am seeing some interesting results. I am seeing a decent amount of newly detected “events”. Whether I am seeing actual events is up to debate. This number depends on what threshold you use. If you use a mean absolute deviation of 9 as a threshold, my template for 2008/05/24, for example, detected 140 new events. However, if you use a cross-correlation value of 0.1 or greater, the value goes down to 40, and if you use a cross-correlation value of 0.2 or greater, the value reduces to 9.
Another issue with the MFT I have come across is deciding which events to use as templates. The first criteria is that the template must come from the same spatial region as the area of study. To find these events, I use a catalog of known events that have occurred in the region. Because I am applying a bandpass filter to my data, distance from the station is also an issue. To ensure that events we are seeing are local (high frequency) events, we only want to use events in the catalog that relatively close to the stations. Ideally, events less than 100 kilometers away from the stations would be used. However, none of the events listed in the catalog for the northeast region of China fit this criteria, so I am using events that are as big and close as possible.
One of the goals of my project has been to see whether or not these newly detected events can serve as evidence for remote triggering. However, we are suspecting that my detections are not identifying enough events to serve as evidence for a significant increase in seismicity. So in the last leg, I will be shifting my focus. A previous intern went through this same region and hand-picked events to identify more than listed in the catalog. So I will be more focused on examining the robustness of the MFT in detecting new events over human detection, and how more efficient it may or may not be.
The past couple of weeks also saw us travelling across China to several other cities. Our first stop was Shanghai. Shanghai is the largest city in China, at least by population and it really shows. There was a large commercialized area with flashy shops that was more crowded than anything I had ever seen. It was so crowded that there were lines to even walk on the sidewalk in some places. But the Bund, Shanghai’s famous skyline, was the most impressive thing I have ever seen in a city. It was sprawling, bright and vibrant. It was never-ending, with no stop to the towering skyscrapers. What was mind blowing to me was that the majority of the skyline was built up only in the last couple of decades.
One thing that has always impressed me here is the stark and sometimes jarring transition from the new to the old here in China. In Shanghai for example, we traveled maybe 10 minutes past the scrawling skyscrapers of downtown Shanghai by train and I was surprised to see many rice fields, even within city limits. You wouldn’t realize that Pearl Tower was only several kilometers away.
Our next stop was Xi’an, one of the most ancient cities in China and even in the world. Here we saw the City Wall, one of the best preserved historic sites in China. Constructed in the 1300’s much of it is still standing and regularly sees tourists walking across it. Perhaps most impressive, though, was seeing the Terra-cotta Warriors in person. I think my favorite part was not seeing the thousands of reconstructed soldiers lined up, but was seeing the sites that are still left as they were thousands of years ago.
I am currently looking at events in the first region I will be looking at, northeastern China. Below is a map of the region. The black triangles are the seismic stations in the 1A array. The circles represent events that are from the catalog of known earthquakes in the region. The red circles represent events that occurred before the 2008 Wenchuan mainshock, and the blue circles represent events that occurred after the Wenchuan mainshock. Because of the scale of the map, it may be difficult to get a geographical sense of the region. For reference, the stations end right at the border between China and North Korea. I produced this map in GMT.
In order to use the Matched Filter Technique (MFT) I need several templates of events to serve as a “comparison” for the smaller events. This past week I have identified several events in the region that could be used as templates. To do so, I looked through the catalog to identify events in the region, and then I looked at my continuous data to see whether or not these known events were detected by my stations. Dr. Peng helped me to decide on the criteria for these events. First, we filtered out any events that would have been too far away from any of the stations. Then we applied a high pass filter, read the data into SAC, and sorted the events by distance from the source. Ideally we see clear p-wave and s-wave arrivals.
Using a MATLAB script, I have been going through each station and producing spectrograms. These spectrograms will hopefully help me identify possible candidates for triggering or microearthquakes that we cannot detect by high (indicated by spikes in the spectrogram). Below is an example of the spectrograms I have been producing. This data is from May 12th, 2008. Image “a” is the raw data of the vertical component of the data from station NE26. Image “b” is the data with a high-pass filter applied, and image “c” is the spectrogram.
The bulk of my time is spent preparing for and trying to apply the MFT to my data. The goal is to expand upon the earthquake catalog and try to identify events that were previously unnoticed. These newly identified events will hopefully be able to serve as evidence for triggering in the region.
China is amazing! We have been to the Forbidden City, the Great Wall, Behai Park, the Olympic Park, the Hutong, and other places here in Beijing. The graduate students at the Chinese Academy of Sciences are extremely welcoming and have been more than willing to show us around the city. One of them took us to see Jurassic World 2 at a movie theater here (the movie was in English with Chinese subtitles). Beijing is such an ancient city steeped with culture and history. I love the architecture of so many of the buildings here. It is also interesting to see the differences between what Americans and the Chinese consider historical. America is relatively young and any building built in the 1700's or 1800's seems very old to us. However, here, they will tell us that something built in the 1600's is young.
And Beijing is huge. I cannot understate how enormous the city is. What looks like a 20 minute walk on a map will actually take an hour and a half (Harrison and I learned that the hard way).
One thing that I still haven't adjusted to here is the fact that many Chinese people do not drink cold water, not even in restaurants. Restaurants will either just not serve water, or serve it hot. I was definetely dehydrated for a while because I was super thirsty and couldn't really get the cold water that I was used to (you can't drink the tap water here). At our office, they do have faucets that dispense boiling water, for easy tea-making. That is convenient, but to drink water that isn't boiling we have to go to the grocery store and buy a pack of water bottles.
These past couple of weeks have been mostly centered around preparing for our travel to Beijing. For me, this has mostly included preparing and downloading my dataset that I will be using for my project. My project centers around the 2008 Wenchuan earthquake, so I have been downloading data from arrays all over China. I am downloading data from the IRIS DMC using a service called BREQ_FAST. This allows me to specify which arrays, which stations, which channels, and when I want the data from.
Another interesting and useful tool that Dr. Peng has prepared me with is the ability to read SAC files into MATLAB. I also am learning how to generate spectrograms using these waveforms. Additionally, I have learned that I can directly load data into MATLAB without having to download it into my computer first. This is useful for loading single waveforms, but loading a lot of data directly into MATLAB takes a long time and I have discovered that with the large amounts of data with which I am working, it is actually easier to download my files directly to my computer first.
This process has definetely taught me the importance of clearly organizing data. It will definetely make processing later on much easier, especially when writing codes that refer to the locations of the data.
Harrison and I enjoyed our last weekend in Atlanta! We saw the Incredibles 2 on its opening night, and we also decided to tour downtown and visit the World of Coca-Cola.
We have finally made it to China! It was a very long flight, about 13 hours, and unfortunately I did not get a lot of sleep on the plane. On the bright side, I did get to watch about 5 different movies on the in-flight entertainment system.
We are staying in dorms here at the Chinese Academy of Sciences. We are very lucky that the dining hall is attached to our dorm building. Also convenient is the fact that the Geology and Geophysics building is directly next to our dorms. It takes us about 5 minutes to get down from our dorms and into our office.
We met with our mentors that will work with us here in Beijing, Dr. Xiaofeng Liang and Dr. Jing Wu. They both made us feel welcome here and I am looking forward to working with them this summer.
For now, I am working on creating a loop that will create spectrograms of all of my data. I will use this to visually inspect the data for evidence of triggering from the Wenchuan earthquake. However, visually I will not be able to see all of the events that may exist. For this reason, we will be working on creating a local catalog of known earthquakes. From there, I will be able to use the Matched Filter Technique to see if there are more events that may provide evidence for triggering in different regions throughout China.
This week, I tried to make an "elevator speech" to breifly describe my project. Going into it, I thought it would be easy. However, it proved to be more difficult than I had originally thought. It was hard to include everything important without going into too much technical detail. I wanted to describe everything in detail, but realized that this was not reasonable when giving a brief summary.
Even though I was very excited to start my internship, I was nervous going into the orientation week. However, meeting all of the wonderful people definitely assuaged all of my worries. We did many great activities, from installing a seismometer to learning how to navigate SAC. I felt I learned so much throughout the jam-packed week with all of the lectures, activities, snack breaks, hikes; I can go on forever about all of the activities that we did to help prepare us for our internships.
On June 2nd, all the interns left to go to their respective destinations. It was a bittersweet moment. On one hand, I felt I had really gotten to know some really amazing people and would definitely miss them until I saw them again at AGU. On the other hand, I was super excited to roll up my sleeves and get started on my project.
I am super lucky to have another intern, Harrison, working on the same project as I am. On that Saturday, even though we had different flights to Atlanta, we met up at the Atlanta airport to grab a quick bite to eat. Then we took an Uber to the apartment at which we were staying. Our Uber driver was extremely friendly and definitely gave off a good first impression of Atlanta.
On Monday, we came into work to meet our mentor, Dr. Peng, for the first time. Dr. Peng is an incredible mentor, and I can tell how extremely knowledgeable about his field he is. It has been a week and I have already learned a tremendous amount from him. I am super excited for the rest of the summer!
This first week has mainly been learning to use the different tools that I will be needing for the rest of the summer. We have been learning how to use Matlab, UNIX, SAC, and GMT. GMT seems to be a very powerful tool for mapping. Dr. Peng has many great tools available on his website that we have used to learn how to navigate several different programs. I really found them well written, but Dr. Peng was still more than happy to answer any questions that we had.
I am very happy to be a part of this project! It is a very exciting process to be a part of, especially exploring new ways to analyze and detect earthquakes. I am also excited to see how microseismicity can lead to our understanding of fault zone structures. In several regions, there is evidence for not only foreshocks, but increased seismicity even days before a larger, mainshock occurs. However, this is often only seen in shallow subduction zones. Some studies have suggested that this increased seismicity can possibly be used as an indicator for larger events.
However, microearthquakes, I have learned, can be useful in studying the spatio-temporal evolution of earthquakes. Main, early events happen very close to or on a fault. Over time, seismicity actually spacially migrates out away from a fault. Seimsologists think this is related to continuous fault slip, and relates to the concept of remote triggering (seismicity caused by larger events that occurred elsewhere). One of the main problems with remote triggering is finding strong evidence relating small events to a large, distant earthquake. This is one of the main topics that I will be focusing on this summer.
In my project description, I describe the Matched Filter Technique (MFT). This week I have learned that there are several limitations to this method:
1) MFT requires a template. That is, if no events from that region have been analyzed, there will be no template and therefore no MFT.
2) Computing may take a very large amount of time. With enough data, this could take days.
3) There is an increasing amount of seismic data being recorded. In order to use the MFT in a timely fashion, we have to, sometimes drastically, reduce the number of waveforms we actually look at.
One of the main challenges that we will be dealing with this summer is getting around these limitations.
This weekend, we decided to explore a little bit of Atlanta! Yesterday we went to the Georgia Aquarium, the largest aquarium in the nation. In fact, they are the only facility in the nation that has whale sharks! It was extremely well-designed and beautiful. One of my favorite exhibits was the large tank with 3 whale sharks, but my favorite part about the exhibit was a particular manta ray. It was fun to watch it, because it kept doing flips in the water. All of the other animals were just swimming in a large circle around the perimeter of the tank, but this manta ray was just having fun doing flips in the middle of the tank for no apparent reason. I also loved the sea lions! We saw a show that the animals put on with their trainers, and I am just shocked about the variety of tricks that they knew. I can't even get my dog to sit, but they could get the sea lions to wave to the audience with a simple hand motion.