Research Experiences for Undergraduates (REU)

Applications for the summer of 2023 will be available starting in November 2022. Please come back then to appy!

Haystack REU students at AGU conference, 2019
Haystack REU students at AGU conference, 2019

Haystack Observatory invites all interested undergraduate students to apply for our paid summer research positions in science, engineering, and computer science. Our REU program has been held for decades, and we have seen many of our student interns go on to rewarding careers in STEM research.

The program extends from early June to mid August. People from groups under-represented in STEM fields are encouraged to apply. Undergraduate students eligible for this program must not have graduated prior to the start of the summer internship in June.

REU and RET, 2018

Details on the program, including schedule, stipend, housing, and transportation are available below.

Support is provided by the National Science Foundation’s Research Experiences for Undergraduates program. The National Science Foundation, which sponsors this program, requires U.S. citizenship or permanent residency to qualify for positions supported under the REU Program. Undergraduate students eligible for this program must not have graduated prior to the start of the summer internship in June.

MIT is an equal opportunity/affirmative action employer.

LEGO: Why Is HCN Unexpectedly Bright in Gas of Low Density?
Anna Apilado, presentation

To resolve the challenges that current observational constraints pose in star formation research, the Line Emission as a Tool for Galaxy Observations (LEGO) project utilizes molecular line emissions to map out molecular distributions within clouds in the Milky Way. As part of furthering the current understanding on star formation, assumptions made about star formation need to be verified. It is a general assumption that (1) star formation rate M is proportional to mass of dense gas Mdg and (2) star formation rate is proportional to luminosity in infrared LIR. When explored whether these assumptions held true, Gao & Solomon reasoned that Mdg is proportional to infrared luminosity of HCN LHCN due to HCN’s high critical density with the common assumption that densities ≫ 104cm−3 are needed to excite HCN. However, the LEGO Survey found that HCN traces densities one to two orders of magnitude less than what Gao & Solomon estimated. This poses the question of why HCN is unexpectedly bright in gas of low density. Possible explanations for this include a higher HCN abundance than expected or an excitation is stronger than predicted. To probe this question, we applied a non-LTE radiative transfer software to measure column density of HCN (J=1-0), corresponded it with the observed brightness temperature from the LEGO survey, and calculated HCN abundance (NHCN/NH2). Input parameters of the software, particularly line width, column density, kinetic gas temperature, and gas density were explored in depth in addition to this calculation.

Automated Detection System for Gravity Waves on Antarctic Ice Shelves Using Supervised Panoptic Spectrogram Segmentation
Shivansh Baveja, poster

The ice shelves fringing the Antarctic continent play a pivotal role in stabilizing the Antarctic Ice Sheet by restraining, buttressing, and modulating the flow of grounded ice into the Southern Ocean. Recent ice shelf collapses suggest that low-frequency (0-70 mHz) gravity wave forcings may trigger rapid shelf disintegration. This study focuses on applying machine learning to automatically detect, classify, and catalog low-frequency gravity wave events impacting the Ross Ice Shelf (RIS) with panoptic seismic spectrogram segmentation. The data used to supervise training were collected by a broadband seismic array deployed on the RIS from November 2014 to November 2016. Our modified U-Net architecture achieved a Dice similarity coefficient (DSC) of over 0.73 during event detection and an accuracy of 94.4% during classification, outperforming alternative rule-based techniques. This work serves as a proof-of-concept for using deep-learning algorithms to detect and catalog gravity wave events, enabling further analysis into the long-term stability of Antarctic ice shelves.

Removing Radio Frequency Interference from Auroral Kilometric Radiation with Stacked Convolutional Denoising Autoencoders
Allen Chang, presentation

The AERO/VISTA missions are twin 6U CubeSat missions that will investigate Auroral Kilometric Radiation (AKR), the strongest of many types of radio emission that originate from energetic electrons in the Earth’s auroral zones. AKR data, typically visualized as time-frequency spectrograms, can be corrupted by noise from nearby electronics and power systems, AM stations, and geo-stationary satellites, limiting the ability to analyze AKR observations. Thus, image processing methods that can reconstruct detail from noisy or occluded observations are a critical pre-processing stage to improve downstream analysis of AKR. We present the Denoising Autoencoder for Auroral Radio Emissions (DAARE), which leverages stacked autoencoders trained with synthetic spectrograms to remove Radio Frequency Interference (RFI) from AKR spectrograms collected at the South Pole Station. We compare DAARE across filtering and deep-learning denoising algorithms with Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity (SSIM) evaluations. DAARE achieves 0.422 PSNR and 0.981 SSIM scores on synthesized AKR observations, improving PSNR by 0.039 and SSIM by 0.064 compared to baseline methods. Removing RFI from AKR observations could assist space scientists in identifying AKR and other auroral emissions from the AERO/VISTA missions. The framework for simulating AKR, training DAARE, and using DAARE can be accessed at

Observing Black Holes with the Event Horizon Telescope
Katy Hunter, presentation

Context Using very-long baseline interferometry (VLBI), the shadow of the M87 supermassive black hole has been imaged by the Event Horizon Telescope (EHT). M87 is predicted to also feature an unstable circular orbit of light called the photon ring outside of this shadow and an inner ring structure created by a relativistic jet expelling ionized matter. While the shadow and outer ring are shifted as the spin of a black hole increases due to gravitational lensing and rotational frame-dragging, the position of this inner ring is not affected by the black hole’s spin.

Aims By measuring the offset between the positions of the inner and outer rings, the spin of M87 can be measured.

Methods This paper simulates observations to test whether the inner and outer rings can be detected at 290 GHz and 345 GHz with the EHT and various ground and space elements, using both SMILI and EHT-imaging techniques. In addition, Variational Image Domain Analysis (VIDA) and Comrade modeling were tested in order to determine whether the offset between the inner and outer ring could be quantified with feature extraction.

Results After determining that the maximum baseline limits of ground VLBI make imaging and modeling the rings impossible from only ground-based telescope arrays, space-based telescope arrays with multiple geosynchronous (GEO) satellites or a single medium-earth orbit (MEO) satellite produced images and models with both rings visible. The offsets per spin were plotted against predicted values for various telescope arrays and modeling methods, and a promising correlation was demonstrated.

Conclusions Although additional iterations need to be performed to verify these results, our experiments support the need, viability, and importance of incorporating GEO or MEO satellites into the EHT array in order to study the structures and spin of M87.

AERO-VISTA Satellites
Alexis Lupo, presentation

Auroral Emission Radio Observer (AERO) and Vector Interferometry Space Technology using AERO (VISTA) are twin CubeSats that will study auroral radio emissions in the ionosphere near the Earth’s poles from low-Earth orbit. The Earth’s auroral regions contain a space plasma that will strongly affect the impedance of the vector sensor (VS) antenna, causing a change in sensitivity. This project seeks to explore this effect by first calculating the plasma impedance and then modeling its impact on the sensitivity along the spacecraft’s orbit. We then use this improved sensitivity within AERO-VISTA’s larger simulation framework to create synthetic data that is representative of a data collect during the AERO-VISTA mission. The framework developed for this project can be used in operations tests as well as the development of algorithms to measure in-situ plasma parameters.

The Great Haystack RFI Hunt
Brian Malkan, presentation

My project aims at detecting sources of Radio Frequency Interference (RFI) across various frequency ranges. I worked with my mentors Dr. Frank Lind and Dr. Sharanya Srinivas towards the ultimate goal in creating training data for machine learning. To detect RFI I created a python script using GNU-Radio, Python and DigitalRF. The script worked in tangent with my “shoebox device”, which is a portable machine I could carry around the observatory to detect RFI from different sources. Once the data was collected, I fed into a data visualization app, made using Dash and Python. From here I was able to annotate specific peaks and discrepancies, and create some example training data.

Simulation of Radar Meteor Ground Illumination to Optimize the Configuration of the Zephyr Meteor Radar Network
Kathryn Postiglione, poster

Since meteors occur randomly within the sky, it is crucial to deploy meteor radar networks that maximize the ability to observe meteors and derive upper-atmospheric winds. To accomplish this task, we developed a meteor simulation model in Python to calculate received power as a function of location given transmitter and meteor properties. An interactive and real-time updating meteor simulation map was created to fully understand each simulated meteor. This map displays received power for specified longitude and latitude ranges. Adjusting the meteor, transmitter, and receiver variables allows the user to identify trends and find optimal transmitter and receiver locations. Since there are an unlimited number of unique meteors, simulations were run to create thousands of distinct meteors for each receiver location within specified longitude and latitude ranges. A special focus was placed on finding an ideal location in Colorado for the Zephyr Meteor Radar Network’s transmitters and receivers, so all simulations were based on longitudes and latitudes within Colorado. Ultimately, the improved placement of transmitters and receivers will result in increased sky coverage of potential meteors and their specular trails, which act as natural tracers of upper-atmospheric winds.

VLA Observation of Teegarden’s Stars
Angelu Ramos, poster

We investigate Teegarden’s Star (also known as GAT 1370, SO J025300.5+165258) for radio emissions caused by magnetic interactions between the star and its planet. This is 1 of 5 stars to
be investigated in a survey that consists of nearby stars that are < 5 pc away. Teegarden’s star is a late M dwarf suspected of being a young star with a planet that is thought to be ~4.88 MNEP with a ~2 day orbit (a = 0.0014 AU). Being an M dwarf star, it has a strong magnetic field but the magnetic field of its suspected planet is unknown which is where radio observations come into play. Planetary radio emissions will be evident through excited charged particles in the planet’s atmosphere, and/or planetary magnetic field interactions with solar wind. Observational data from the VLA in the P Band (0.23 – 0.47 GHz; 90 cm) and Teegarden’s star strong magnetic field allows for reasonable ground based survey. The VLA observed Teegarden’s star on August 16, 2016 within the times of 11:31:50 to 12:11:40 (~40 minutes) on the same day. With NRAO’s Common Astronomy Software Applications (CASA), we are able to calibrate and image the P Band data – imaging is taken a step further by imaging between the different spectral windows and then by time frames. In our process, we did not find any radio emissions from Teegarden’s star but have established an upper limit on P-band radio flux from this system.

AERO-VISTA Satellites
Max Riccioli, presentation

The AERO-VISTA mission will study auroral radio emissions and demonstrate interferometry using electromagnetic vector sensors from low earth orbit through eponymous twin 6U cubesats Auroral Emission Radio Observer (AERO) and Vector Interferometry Space Technology using AERO (VISTA). As part of this mission, a reliable and extendable software package has been developed to facilitate communication between the satellites and ground stations. This project has added critical improvements to the core ground station software: the ground station interface is now capable of managing several GNURadio flowgraphs concurrently, as well as loading arbitrary flowgraphs from file; configuration options have been extended for many components of the ground system; mission-specific definitions, routines, and flowgraphs have been reorganized and consolidated into a central package. Additionally, numerous flowgraph-level enhancements have been made: defective packets can be detected in-flowgraph, preventing packet drops due to apparent overlap; improved demodulation and signal processing methods reduce GFSK bit-data quantization errors; and data pre- and post-processing methods reliably discern packet transmissions from radio noise. Finally, the ground station interface’s existing Redis server integration has been leveraged to send uplink and downlink entries across NNG pipes, allowing for eventual connection to NanoMCS for complete data packetization and depacketization as will be seen during actual mission operation. Before the package is mission-ready, further integration with other mission-level software is still necessary, as are more robust signal processing options. With the above improvements, however, a flexible framework has been created to allow for quick development of these features.

Resolving the Motion of the Accretion Flow and Jet of M87*
Audrija Sarkar, presentation

The Event Horizon Telescope (EHT) currently produces static images of the emission surrounding supermassive black holes. A key goal over the next few years is to lengthen the observing window in order to produce videos of the supermassive black hole at the center of Messier 87, M87*. An important purpose of such a video would be to resolve the complex dynamics of material in the accretion flow and jet. In this project, we have developed a tool to extract and characterize motions. For the motion extractions, two methods were examined using videos generated from general relativistic magnetohydrodynamic (GRMHD) simulations: voxelmorph, a neural network used in medical imaging, and Lucas-Kanade method in opencv, a classical technique to measure the dense optical flow. We find that the Lucas-Kanade method is more effective at detecting spiral motion than voxelmorph. We further characterized the extracted angles of rotation and velocity field with polar Fourier analysis. These results demonstrate that optical flow methods may be used to characterize the dynamics of the accretion flow and jet from EHT observation.

Statistical Studies of Traveling Ionospheric Disturbances and Equatorial Plasma Irregularities
Tal Sternberg, poster

Travelling ionospheric disturbances (TIDs) occur as a result of space and terrestrial weather. Some previous case studies have suggested TIDs are likely associated with the presence of equatorial plasma irregularities (EPIs). This study provides statistical analysis of both TIDs and EPIs in the America sector as well as their potential correlation. The GNSS measurement of total electron content (TEC) from ground-based receiver networks can be used to identify potential TIDs. A large global TID dataset based on ∼6000 GNSS receivers is being created at MIT Haystack Observatory for every day since 2018 using a previously developed TID detection algorithm where a 30-min sliding widow is used to detrend background ionospheric variations. Monthly averages of TID amplitudes with up to 1 deg latitude by 1 deg longitude by 5 min time interval spatial-temporal resolution, are calculated. These allow us to examine the TID climatology on local, regional, and global scales. We have developed various visualization tools to assist in the characterization of TID climatology including day-to-day, seasonal, and long-term variations. Equatorial Plasma Irregularities (EPIs) are normally considered to be developed under the Rayleigh-Taylor instability where the seeding effect of neutral and ionospheric waves is important. To characterize EPIs, we use the GOLD’s nightglow measurements at 135.6 nm. A statistical analysis for the EPI occurrence in the GOLD data is performed over the years since 2018, corresponding to the period when the GNSS TID climatology is derived. In this presentation, we will report climatological results from both TID and EPI statistical studies, and explore the potential connection of EPI development to TIDs.

Sensing Snow Depth over Arctic Sea Ice Using GPS Reflectometry
Sarah Zhang, poster

The thermal insulating properties of snow are quite extraordinary. In the Arctic Ocean, processes, such as sea ice formation and melt, are partly regulated by the thickness of the snow layer that overlies the sea ice. It is therefore important to obtain reliable estimates of snow depth over sea ice to understand ongoing changes in Arctic Ocean sea ice cover. Here, we explore the feasibility of snow depth estimation using the Global Navigation Satellite System-Interferometric Reflectometry (GNSS-IR) technique. This remote sensing technique uses the interferometric pattern in the GNSS Signal-to-Noise Ratio (SNR) observable, which results from the interference of the direct and reflected GNSS signal, to estimate the vertical distance between the receiving GNSS antenna and the reflecting surface. In the case of Arctic sea ice, the reflector is the snow, and vertical distance changes are due to snow accumulation or melt. The cm-level precision of the GNSS-IR technique has previously been demonstrated using stationary GNSS antennas. Here, data from the Sea Ice Dynamic Experiment (SIDEx) is used to investigate the precision of kinematic GNSS: GNSS antennas are anchored to an ice floe, drifting with the Arctic Ocean ice pack. We have processed approximately one month of GNSS data from 12 identical GNSS systems deployed during the March 2021 SIDEx campaign, forming a small-scale network of ~5 km. There are noticeable differences between systems, possibly attributed to the quality of the reflecting environment. Overall, the cm-level precision of GNSS height estimates in this kinematic environment demonstrates consistency with the precision of estimates from static GNSS-IR studies. These are promising results for validating the most recent estimates of snow depths from remote sensing missions, such as ICESat-2 and CryoSat-2.

REU summer projects from past years are available in the presentation archives.

Program details

Please see the following sections for general information about the Haystack REU program. (Note that all information on this page is subject to change if necessary.)

Q: When can I apply for next summer? What is the time frame for applications?

A: Applications are made available on Thanksgiving each year for the following summer. The deadline is February 1; we notify successful applicants on March 1, followed by a series of notifications for acceptances if any are open after the first round of notifications. Please do not contact us about the status of your application. If you are not accepted, you will receive notification of this in March after all positions have been filled.

Q: My institution is on a quarterly system (or another schedule). This means that I won’t be available on the exact start date. Can I still apply?

A: Yes, please apply to the program. Make a note of your possible start date in your statement letter. The mentors will determine whether your earliest possible start date is acceptable within the requirements for their project. We prefer that people be available for the actual start date but realize that some institutions’ schedules make this difficult.

Q: Do you accept international students?

A: Unfortunately our sponsor, the NSF, requires applicants to be a U.S. citizen or permanent resident to qualify. We won’t be able to respond to inquiries regarding this.

Q: How is “undergraduate student” defined for this application?

A: An undergraduate student is a student who is enrolled in a degree program (part-time or full-time) leading to a baccalaureate or associate degree. Students who are transferring from one college or university to another and are enrolled at neither institution during the intervening summer may participate. High school graduates who have been accepted at an undergraduate institution but who have not yet started their undergraduate study are also eligible to participate. Students who have received their bachelor’s degrees and are no longer enrolled as undergraduates are not eligible to participate. Undergraduate students eligible for this program must not have graduated prior to the start of the summer internship in June.

Q: Will the REU program be held in person or online for 2022?

A: We hope that the Haystack internship program will be held in person in summer 2022. However, this decision is affected by local and MIT safety and health regulations. We continue to monitor the situation closely. (In 2020 and 2021, our research internship program was successfully held completely remote.) More to follow as information becomes available.

Q: Do you send notification letters when my application is submitted?

A: We will notify everyone of their application status after the final deadline for submission (February 1) has passed. If you are missing a piece of your application, we’ll let you know then.

Q: I applied but did not hear from you on March 1. How will I find out about my application’s status?

A: The first round of acceptances is sent out every year on March 1; if any of these positions is not accepted, it will be offered to the next round of candidates on March 8, and so on until all of the positions are filled. This means there is a series of acceptance letters, starting on March 1 and possibly continuing into March. We do not inform applicants of their status until all the positions have been accepted; if you are offered another position in the meantime, we suggest that you accept it, as we have only a very limited number of internships available.

The Haystack REU program starts in early June and ends mid-August. The 2022 REU program at Haystack will run from June 5, 2022, through August 12, 2022.

A uniform start date is preferred in order to conduct orientation activities for the group. For students on an academic quarter system or those interested in extending their stay, such requests can be considered on a case-by-case basis.

The Haystack summer undergraduate internship includes participation in the following:

  • Science discussions: Haystack staff members lead discussions on numerous current research subjects, which include introductory information for all students, as well as a chance for active conversation with scientists, engineers, and other staff.
  • Tours: Students will attend tours of the various Observatory facilities to learn about the extensive state-of-the-art instrumentation at Haystack. 
  • Group meetings: In addition to the frequent meetings between the sponsoring staff member and the student, several meetings with all students are held to review project status and encourage interactions among the students.
  • Final reports and seminar: Students prepare brief final reports and create presentations on their projects to teach the Haystack community about their summer work.
  • Attendance at conferences: Depending on available funds and meeting schedules, there are opportunities for students to participate in national conferences.
  • Follow-up academic year program: Depending on available funds, interest, and project status, a student may continue the summer project during the following academic year.
  • Travel support: Limited travel support is available for those students whose homes and colleges are more than 100 miles away from Haystack.

Students are assigned a mentor from the Haystack research staff for their summer work.

At the end of the summer, students present their research to a general audience at the Observatory. Their presentations are available in the REU presentation archives.

Compensation is provided as a weekly stipend of $600, paid biweekly.

The Observatory makes arrangements for student dormitory housing and pays the cost of housing for all students. Kitchen facilities are available in the dormitories. Daily transportation to and from Haystack is also provided.

(Students can arrange alternative housing on their own if they wish.) 

The Observatory provides free daily transportation for all students from REU housing to our offices.

Accepted students must have a current medical insurance plan in place which will cover their health needs during the period of the REU program. Evidence of such insurance must be submitted upon acceptance, before the start of the program.