Seismology has historically been a key method for verifying nuclear test detonations. A nuclear detonation's characteristic seismic "signature" is distinct from that of an earthquake; based upon seismic data from numerous instances of either event, a dividing line has been empirically established. This line demarcates earthquakes, which usually generate strong shearing S-waves, from detonations, whose energy is mostly converted into heat and compressional P-waves. However, recent nuclear test detonations executed by the Democratic People's Republic of Korea (DPRK) are poorly characterized by this model, leading to concern over "false negatives" and highlighting the need for a more physics-based event differentiation model. The National Nuclear Security Administration (NNSA) is seeking to create such a model using data from their Source Physics Experiment (SPE). SPE consists of a series of chemical detonations of varying yields emplaced in different media at various depths. SPE shots are executed at the Nevada National Security Site (NNSS, formerly Nevada Test Site), an area where over 900 nuclear detonations occurred between 1951 and 1992. The data from SPE will be used to build a more reliable and physics-based model that more thoroughly accounts for source and medium properties. Recently, project scientists determined that imaging the subsurface of the explosion area prior to the next series of SPE shots would better inform the physical models applied to the SPE data. In 2015, members of a team from Sandia National Laboratories, under project THOR, acquired exquisite active-source seismic data for Yucca Flat, Nevada, the chosen location for the next phase of SPE shots. The source was a 13,000-kilogram modified industrial pile driver called the Seismic Hammer™. This summer I will be performing surface-wave group velocity tomography on THOR data to construct a shear-wave velocity model of the subsurface of Yucca Flat. Surface-wave phase velocity tomography was previously employed in an attempt to obtain a shear-wave velocity model; the results of this approach were significantly distorted in areas where seismic ray paths were proximate to former nuclear detonation sites, resulting in an incomplete velocity model. We hope to obtain higher-quality results using group velocities, leveraging the assumption that - compared to individual phases - the propagation of wave packets is less impacted by the fractured, disrupted media surrounding former detonation areas. The resulting shear-wave velocity model will be integrated with other tomographic results to create a complete model of the subsurface structure of Yucca Flat that can be utilized in the physical modeling of SPE data.
That’s a wrap! My nine-week internship at Sandia has come to an end. Looking back at the goals I set during week one, I can safely announce that I’ve achieved about a solid 60% of them. Hey, that might actually be better than an “F” letter grade, completion-wise! Yay.
What’s much more important to me is the shift in my outlook regarding graduate school, a change induced by this internship. Heading into this experience, I was skeptical of the idea of staying in school for at least two (and maybe longer than seven!) more years. After my time at Sandia, however, I’m convinced that my desire to continue engaging with seismology will only be satisfied by graduate study of some sort. My current plan is to take a “gap” year, during which I will still study geophysics – albeit outside of the strictly academic context. Then I’ll enter into graduate school the subsequent year. I’m enthusiastic about this plan, since it allows me a break from academic structure while still keeping me on track for a potential career in geophysics.
I’m going to sign off for the time being. Thanks for reading. Maybe there will be another blog post here in December on the topic of AGU Fall Meeting adventures. If not, I’d like to use this post as an opportunity to thank all of the folks (Michael Hubenthal, Rob Anthony, Hunter Knox, Rob Abbott, and many more) who played a role in this terrific experience. It has meant a great deal to me, and I hope to give back in a similar manner later on in my career. That’s all – Liam out!
Well, that was quite the week. Rebekah and I departed ABQ on Sunday in the early morning, driving eight and a half hours out to North Las Vegas (NLV), our home base for the duration of the experiment. The drive, which normally would have been fairly uneventful, was made a bit more exciting by the massive vehicle we were piloting. Nicknamed “the A-team van,” it’s a lifted 4x4 E-Series van with a pop-up top and massive tires. It certainly drew some attention at gas stations…
Figure 15: We may have accidentally run over some Honda Civics on our way to Vegas.
During our week of fieldwork, the schedule generally proceeded as follows: Meet in the hotel lobby at 5:15am; arrive at the Nevada National Security Site (NNSS) at 6:30; reach our survey location at 7:30; work until 4pm; arrive back in NLV at 6; grab dinner somewhere and pick up lunch for the next day. Needless to say, it was exhausting – the 100 °F temperatures at the survey location didn’t help, either.
The cross-borehole survey we conducted required the use of two sources: An accelerated weight drop (AWD), mounted to the hitch of a truck, and a high-voltage sparker, which we dropped into the water-filled boreholes. For our receivers, we utilized a line of hydrophones (essentially microphones for water) as well as conventional single- and three-component geophones. As you may have guessed, the sparker generates a spark. This spark plasmatizes the surrounding water and creates a large bubble, which impacts the wall of the borehole, creating a seismic wave. The sparker was controlled by a trigger box, upon which was mounted a large red button. During several extended intervals of the field test, my sole job was to listen for the command to fire, and then press the button. Again. And again. And again. The text message exchange between Rob and me, shown below, is a good representation of my outlook during those extended periods of button-pressing.
Figure 16: Rob, showing great sympathy.
By Friday afternoon, we had acquired all the data that we needed (and had also depleted an entire 5,000-gallon water truck as we refilled the constantly draining boreholes). Rebekah and I drove home on Saturday and managed to finish the entire audiobook for The Martian by Andy Weir. Highly recommended. In spite of our intense exhaustion, we all made it out to an intern BBQ at Hunter’s house, which was great fun. It’s probably pretty easy to gauge my level of fatigue from my facial expression in the photo below.
Figure 17: Three generations of IRIS interns! From left to right: Tori, Max, Kevin, me, Olivia, and Rebekah.
It’s hard to believe, but next week is the last week of my internship at Sandia. There are a lot of loose ends to tie up. I’ve also been discussing post-undergraduate paths with Hunter, Rob, and others. There’s quite a bit to consider, but I’m becoming increasingly amazed at the diverse array of prospects!
This week I checked over my newly filtered data and ensured that my filter script hadn’t made any mistakes. Then the next task was to create a “pickfile” – the plain text file that our tomography code takes as an input. The “pick” of a trace, in the context of our tomography process, is the timestamp of the peak amplitude of that trace, which we then convert to velocity since we know the source-receiver offset. This text file is huge. Remember those roughly 1.3 million computations that I had to perform? Well, they translate into an equal number of picks: My text file contains over 1.3 million lines. I guess I’ll have an easy way out if my senior thesis is under the required page limit…
I’m supposed to outline a few challenges I’ve encountered during the internship, so here goes. I’m a perfectionist, which works with me or against me depending upon the task at hand. For example, I create countless in figures in MATLAB. Only a choice few of these are designated for inclusion in my AGU poster; most are simply QC (quality control) diagrams that needn’t be highly polished. Too often, I waste time producing a pretty diagram that only I (or maybe Rob) will ever see. I’m finally learning how to be judicious in my use of time when it comes to refining the content I create, identifying what’s a publication- or presentation-level figure versus an intermediary processing tool.
Another challenge I’ve faced is grasping the big picture. It’s dangerously easy to refine a particular processing method for a single source-receiver pair, only to realize that such a method doesn’t work on a broader scale! I’ve been guilty of performing plenty of my QC’ing on only a few select traces. Now, I’m trying to utilize tools such as shot gathers and multi-shot scatter plots to determine if my processing methods are working as they should. This requires more effort and a markedly different approach, but looking at the big picture frequently will prevent me from developing a processing method that is “hard-coded” to only a tiny sample of my massive dataset. I’ll try to make this a habit from now on.
At the end of the week, I joined Max and Olivia for a whirlwind tour of Carlsbad Caverns National Park and White Sands National Monument. The caves were impressive – they went on for what seemed like forever, covered with intricate features, and the temperature was a cool 55 °F inside.
Figure 13: Beautiful “decoration” in the caverns.
The heat at White Sands provided a searing contrast to the coolness of the caverns. Naturally, we were there during the hottest part of the day, and it was well over 100 °F. We experimented with dune sledding, and managed to avoid becoming lost in the sea of identical white mounds. Fun fact: The dune sand in the area is composed of gypsum, which makes it much less susceptible to solar heating than quartz sand. Even in the middle of the afternoon on a clear, sunny day, we could frolic around barefoot and be perfectly comfortable. Yay science!
Figure 14: Enjoying the dunes. Photo credit to Olivia.
This week, I learned the importance of running one’s hefty MATLAB scripts over the weekend whenever possible. Rob and I recently made the decision to filter the THOR data using a center frequency range of 1-50 Hz instead of 1-100 Hz. Since my Gaussian band-pass filters require equal relative bandwidths, this involved re-designing my set of 30 filters and subsequently re-filtering the data. This last step is what taught me my lesson. I had over 150 shots to filter. Each contained about 300 to 350 traces. Each trace had to be filtered 30 times. That’s at minimum over 1.3 million operations. To compound everything, I was using
Long story short, my poor Mac mini kept a high temperature most of the week as it churned through trace after trace (unfortunately, no faster machine was available). The whole process ended up taking over 80 hours to complete. While the filter script was chugging along, I began tackling the rather steep learning curve of the Generic Mapping Tools (GMT), a set of command-line tools used to make manuscript-quality plots and maps. After quite a few frustrating hours, I ended up with this GROOVY map of Yucca Flat.
Figure 11: A map of the THOR study area, composed in GMT. White circles indicate shot locations. Check out all those craters!
After scrambling past the steepest part of GMT’s learning curve, I can certainly see the benefits. One can make simple maps with just a few lines of code, and it’s free, which is a godsend considering how expensive licenses for dedicated software packages such as ArcGIS can become. GMT is quite popular with geophysicists for these reasons. Plus, with all the funky option flags and lack of spaces, it makes you feel like a super cool coding expert when you get things right. That’s probably another reason why geophysicists favor it…
I’m eagerly anticipating our upcoming seismic data acquisition campaign at the NNSS. There’s been a seemingly never-ending stream of training and logistics to work through, and I’ve already learned loads about the intricate preparation required for a field experiment taking place at such a secure location. I’m especially looking forward to the change of scenery. After weeks sitting at a computer, spending some time outside performing physical labor will be a welcome change – even if it’s going to be tough.
Outside of the science realm: This Saturday I realized – unbearably late into my internship – that ABQ has its own lift-served bike terrain! There aren’t too many trails to explore, but nevertheless it’s always fun to ride on a chairlift in the summer, especially when it’s in lieu of climbing thousands of feet on your bike at high elevation. A brief hike got me from the top of the lift to the crest of the Sandias, which presented a breathtaking view. The east side of the Sandia crest is fairly gentle (that’s where the ski and biking area is) while the west face, the part visible from the city, is aggressively steep and rugged. The drop-off is nearly nausea-inducing, and the city looks strange, almost too flat, from the high vantage point. I’d love to return at some point and catch the sunset, since I’m convinced it would be absurdly gorgeous from the crest.
Figure 12: After a long day of descending.
Amidst a frenzy of data processing, background reading, mountain biking, brewpub sampling, and blogging (amongst other activities) the halfway point of my time at Sandia has flown by! Thankfully, progress has been made, and Rob and I are performing the final decision-making on what parts of the data to include in the tomographic analysis. One of the key plots we’re using to pick out Raleigh wave arrivals from other seismic signals is called a hodogram (not to be confused with hodorgram, in spite of the obvious similarities).
Figure 9: A hodogram displaying the retrograde elliptical motion characteristic of Rayleigh surface waves.
One can think of a hodogram as a plot displaying the motion of an individual ground particle over time as a seismic wave propagates through the medium surrounding that particle. Since different seismic waves influence ground particles in unique ways, hodograms are a great tool for discerning between various wave types. We’ve used hodograms to confirm that the energy in the vicinity of 2.8 km/s discovered previously does indeed correspond to a ground-coupled air wave – the hodogram of that region exhibits prograde particle motion, essentially precluding the presence of a Raleigh wave. Even though they’re fairly straightforward to create, hodograms are probably my favorite plot type out of the various plots I’ve created thus far. They’re an extremely simple and elegant – yet powerful – way to visualize a wave’s motion.
Big news came this week: Hunter has informed me that I’ll be travelling to the NNSS (formerly NTS, the Nevada Test Site) for a week in late July to conduct some cross-borehole fieldwork. Though I don’t know much about the nature of the experiment we’ll be executing, Hunter assures me that it is very hard work. Still, I’m unbelievably excited to get outside and do something “hands-on,” especially in an area that so few people get to see!
To add to my growing exhibit of adventure photos, here’s a snap from my Saturday bike trip to White Ridge, northwest of ABQ. The trail was pretty narrow and exposed in some places, and the scenery was unreal – a potentially hazardous combination! It was another phenomenal day of riding. I could get used to this.
Figure 10: Looking over the “Dragon’s Back” at White Ridge. This trail demanded focus!
After spending over a month working at Sandia, I’m still regularly fascinated by the culture and environment. There are certainly pluses and minuses to working at a national laboratory, but either way it is most definitely interesting. Since Sandia is located on Kirtland Air Force Base, I’m required to show a base pass and photo identification to an armed serviceperson in order to drive onto the base. Then I drive to my Sandia building; I swipe a badge to get inside.
Once this whole fairly serious process is complete, however, actually working inside the building is surprisingly normal. I'm stationed in a room with a total of about 10 other interns – we’re a mix of geophysics, atmospheric sciences, and engineering students, all collaborating with various mentors. Interning at Sandia can be quite humbling, since there are so many subject matter experts here; truly some of the best of the best, in all sorts of arcane fields. But that is certainly an advantage whenever I need assistance with a science problem!
|Figure 6: My lovely workstation, including an underpowered but trusty Mac mini.||Figure 7: The geophysics crew: Charles, Stephanie, and (to the right of my desk) Rebekah, the IRIS intern at Sandia two years ago!|
Speaking of science, Rob and I have been working on identifying wave arrivals on the dispersion plots I’m creating (see Figure 5). It turns out that the high-amplitude region (red on the plot) at about 8 Hz lines up with a slowness of about 2.8 km/s. This translates to a velocity of a little over 350 m/s – roughly the speed of sound in air at warmer temperatures, like those found at Yucca Flat. The sound waves created when the hammer strikes the desert surface are very strong, and they vibrate the ground near each receiver, causing a signal to appear in the data. This discovery is important to us, since we are primarily concerned with seismic surface waves – not ground-coupled air waves – for our velocity model.
I had a full four days off for the holiday weekend, so I headed up to Grand Junction, Colorado to visit some family. I seized the opportunity to ride some of the world-class trail networks in Grand Junction and nearby Fruita, and I had the privilege of driving through Colorado National Monument, which was very impressive. I could definitely imagine living somewhere in the Southwest later in my life!
Figure 8: A view of Colorado National Monument, Fruita beyond.
This week we were asked to write a brief “elevator pitch” describing our research – something we could give at a poster presentation, an interview, or to a family member or friend. The ability to describe one’s research concisely yet fully is already not an easily acquired skill. Doing the above while also keeping the jargon to a reasonable level (perhaps for a technically-oriented non-scientist) is a skill some scientists – dare I say, professors? – never fully master.
Writing my pitch, I found it very straightforward to talk about the motivation for my research, since it’s quite clear what the underlying purpose is, and the basic politics of nuclear weapon testing are known to most (i.e. it’s “not chill”). The data gathering procedure was similarly easy to describe; it doesn’t get much simpler than “hitting the desert with a freaking massive hammer.” What I found most difficult was explaining exactly how I’ll be using the THOR data to create a model of the subsurface. I’m beginning to realize that this implies I’m not entirely confident in what’s going on myself. My position is best summarized in the below quote, by an obscure scientist you’ve probably never heard of:
“If you can’t explain it to a six-year-old, you don’t understand it yourself.” – Albert Einstein
This idea shares philosophical underpinnings with the notion of Occam’s Razor. Together, these concepts encourage us to seek out the simplest answer, which is normally the one best communicated and understood en masse. A scientist can always complicate their model and add in additional information if the simple construction is not enough. Keeping these approaches in mind, I’m continually re-framing the way I describe my research to others, trying to paint the most concise, complete, and unfussy picture of project THOR.
Speaking of THOR, I have to admit that I’m cheating by showing and discussing a figure that I created this week, the fourth one of my internship, rather than last week. However, it’s a big enough step in the process that it might as well be shown now.
The figure below is a visual representation of what surface wave dispersion looks like; shot 78, trace 153, of the north-south line of the THOR campaign (THOR1) is depicted. This single wiggle was filtered at 30 different frequencies – these are marked on the horizontal axis. On the vertical axis, we have “slowness,” which is the inverse of velocity, in seconds per kilometer. I have to admit I chuckled a bit when Rob introduced me to that particular quantity… The colors indicate signal strength, with hotter hues corresponding to larger envelope amplitudes. Since we are assuming that the peak amplitude of the wave envelope corresponds to a surface-wave arrival, the hottest-colored portion of each column roughly indicates the time at which the surface wave arrived, filtered at a given frequency on the horizontal axis. Since we know the distance between source and receiver, we can transform pick times into velocities or “slownesses.” One can think of this type of figure as 30 “sub-traces,” all filtered at different frequencies, stood on end and concatenated.
Figure 5: Houston, we have dispersion!
The good news is that our plot has the correct trend. We see that lower-frequency waves are “less slow,” or higher velocity, than higher-frequency waves. This is what we expect, since the lower-frequency waves can travel at deeper depths where media are generally faster. Unfortunately, we hand-selected this particular trace since we possessed pre-existing knowledge that the data was robust in the vicinity between source and receiver. If we’re trying to do better with our group-velocity approach compared to last year’s phase-velocity method, we’re going to have to make sense of plots similar to the one above but for traces showing signs of scattering from past detonations. We have our work cut out for us!
My second week at Sandia is over, and I’ve become much more familiar with MATLAB, perhaps at the expense of getting myself more acquainted with the actual THOR data. I’ve spent most of the week working on the design of a highly specific band-pass filter through which I’ll pipe each trace (wiggle) from THOR. The reason we’re implementing such a filter has to do with the unique properties of surface waves. Surface waves of lower frequency are intrinsically less attenuated compared to higher-frequency waves, since they go through less cycles over a given distance. These longer-period waves are also more sensitive to deeper depths, where media are less attenuating. Our method involves band-pass filtering each trace at a multitude of different frequencies, since each particular frequency essentially corresponds to a particular depth.
Though I haven’t delved too deep into the THOR data yet, I know enough about the acquisition process and resulting signal attributes to understand that this is a dataset of superb quality. One mark of a high-quality dataset is a large signal-to-noise ratio (SNR). This is a measure of how clearly seismic wave signals stand out over the background (ambient) noise level. One method for increasing SNR when collecting active-source seismic data is to hit the ground several times at a given location and “stack” (sum) the resulting traces. Since both source and receiver properties are invariant throughout the stack, phase consistency ensures that the peaks and troughs of the actual seismic wave signal constructively interfere, while the random ambient noise roughly cancels out. The result is a “stacked” waveform with a boosted signal and reduced background noise – a very desirable result.
During the THOR acquisition campaign, the Seismic Hammer™ beat the ground more than 30 times at each shot point. This is a massive number of strikes! The resulting stacked traces exhibit a very large SNR; this benefits everyone involved in the post-processing stage. There’s a saying in audio production, meant to convey the importance of a high-quality raw recording: “Garbage in, garbage out.” THOR project scientists certainly took this warning to heart, and the result is a pristine data set which I am very lucky to have the privilege of processing. The THOR data is not publicly available at this time.
This week I learned an important research philosophy, best summarized by the advisory maxim Hunter shared with me after discovering that I’d spent hours building my own Gaussian filter from scratch in MATLAB: “Don’t be a hero,” she said, showing me
Figure 3: MATLAB’s filter design interface, visualizing the frequency response of a Gaussian band-pass filter.
I capped off my four-day work week with a nice suite of outdoor adventures, beginning on Friday with a rocky singletrack ride skirting USAF property (usefully delineated by pleasant “unmarked unexploded ordnance area” signs). The ride included a couple of near-miss falls, one of which involving an extremely unfriendly interaction between a prickly pear and the palm of my hand. I spent Saturday exploring the unique Valles Caldera (photo below) and Bandelier National Monument with Olivia, Kevin, and Max (ASL). I feel extremely lucky to have such a wonderful group of intrepid ABQ-area interns who are so willing to explore what New Mexico has to offer!
Figure 4: A view of Valles Caldera, taken from the highway.
I’ve been in Albuquerque (hereafter ABQ) for just over a week now, and after taking care of a few issues – such as multiple faulty refrigerators, and almost losing my badge on the first day of work – I feel fairly settled. I’ve had time to speak with my mentors, Hunter Knox and Rob Abbott, concerning my project, and that has helped make my summer’s trajectory a bit less nebulous.
I devoted much of this week to the gathering and perusal of readings pertaining to my research topic, including surface waves, dispersion, phase versus group velocities, and depth attenuation. My physics background has helped me navigate through some of the material, but the physics is being applied in a context with which I’m totally unfamiliar, so there’s still a lot to learn. I’m also getting my hands dirty with MATLAB for only the second or third time. Late last week, I loaded the THOR data into MATLAB and practiced executing some simple manipulations. Here’s a wiggle to prove it:
Figure 1: A wiggle.
As this summer starts rolling, I’ve been tasked with defining a set of measurable goals, and an accompanying temporal framework, that will give my summer structure and help ensure a successful outcome. Keeping this in mind, I’ve split the summer into three phases, each with a specific goal or checkpoint.
1st third: Understand the physics of surface wave dispersion; get comfortable manipulating THOR data; use MATLAB to filter the data in preparation for picking.
2nd third: Pick peak amplitudes of wave packets; massage data in preparation for tomographic analysis; perform thorough quality control.
Final third: Execute the tomographic inversion; assess results; prepare AGU abstract and poster.
A large priority of mine (that doesn’t necessarily fit into the timeline above) is to learn the precise language of the field; for instance, instead of referring to a “shear wave that moves up-and-down,” I’ve begun to say “vertically-polarized shear wave” or just “SV-wave.” By refining my vocabulary within the sphere of seismology, I’ll be able to communicate my ideas more effectively during this summer and beyond.
I have a couple of other goals that are unrelated to my life at Sandia. I want to take my mountain bike out at least twice a week, riding local and regional trails. I hope to get out to Sedona, AZ at some point during the summer! I’m also on a mission to spend every weekend at a new location outside. Just this past weekend, I climbed up to Katherine Lake (at over 11,000 feet) with my fellow ABQ-area interns, Olivia (LANL) and Kevin (UNM). I didn’t realize that New Mexico was home to such gorgeous alpine scenery!
Figure 2: Beautiful Katherine Lake, nestled away in the Santa Fe National Forest.
Today is the last full day of orientation week here in Socorro, NM. Over the past five days, we've been inundated with information concerning everything from tomography to networking to spectral analysis to graduate school opportunities. It's shaping up to be a phenomenal summer!