Embedded Files
Contents
How It Started
How It Started
My enthusiasm for studying the Earth started quite early, when I was a teenager. As someone fascinated by both astrophysics and meteorology, it was perhaps natural that I found my way to the Earth Science Olympiad. After a competitive (but also genuinely enjoyable) selection process, I was selected to represent Korea and won a gold medal at the 4th International Earth Science Olympiad in Indonesia. It was an unforgettable experience, and it was around then that I decided I wanted to become a professional researcher.
Building a Foundation in Physics
Building a Foundation in Physics
I studied physics at the Korea Advanced Institute of Science and Technology (KAIST) in Daejeon, a peaceful city in the middle of South Korea. I chose physics because I wanted a rigorous foundation for eventually studying something as complex as the Earth system. Looking back, it was a good decision. In particular, the training in electromagnetism continues to help me enormously in my research today.
KAIST also encouraged undergraduates to explore courses across different departments, which gave me the freedom to study not only physics but also engineering and other subjects. I became especially interested in nonlinear systems. What fascinated me was how closely some of the ideas resembled the Earth system: we can never observe every part of a complex system in complete detail, so we have to understand it through limited observations that encode information about the underlying state.
A year before graduation, I found my way back to Earth science. I had an opportunity to join the Satellite Meteorology Laboratory at Seoul National University (SNU), led by Prof. Byung-Ju Sohn, who later became my PhD supervisor. For my undergraduate graduation project, I spent about a year studying aerosols and haze events over Seoul.
This was my first real encounter with satellite remote sensing, and I was fascinated by its ability to reveal Earth-system variability over enormous spatial scales, including regions that are difficult to reach or monitor continuously. Publishing my first paper, on the characteristics of haze events over Seoul, was therefore particularly meaningful to me. I am especially grateful to Hwan-Jin Song and Young-Chan Noh, who mentored me during this first step into research.
From the Atmosphere to Arctic Sea Ice
From the Atmosphere to Arctic Sea Ice
After graduation, I entered a combined master's–PhD program at SNU. I was fortunate to be surrounded by researchers working on a remarkably broad range of topics, including global and regional precipitation, atmospheric temperature and humidity profiling, numerical weather prediction data assimilation, and radiative transfer modeling.
As a junior graduate student, I spent a productive period learning about these different areas. In that sense, I was initially trained as an atmospheric remote sensing scientist. I also gained a broader view of how the international scientific community works by supporting my supervisor's activities within the International Radiation Commission.
Meanwhile, rapid warming in the Arctic was drawing my attention toward the cryosphere. Researchers were increasingly discussing Arctic amplification, the fact that the Arctic was warming several times faster than the global average, and its connection with the rapid loss of sea ice. I became interested in whether we could produce sufficiently reliable long-term sea-ice records to better understand these changes.
Persistent cloud cover and the absence of sunlight during the polar night naturally led me toward microwave observations. But microwave radiances over sea ice are influenced simultaneously by the atmosphere, snow, ice, and ocean. To interpret them properly, I needed to understand how the physical properties of this coupled system are transformed, through electromagnetic interactions, into the radiances measured from space.
This was also when I began working closely with Sang-Moo Lee, then a senior graduate student and now a professor at SNU. Together, we investigated sea-ice emissivity and the roles of volume and surface scattering in microwave observations (Lee et al., 2018).
Snow, Snow, Snow!
Snow, Snow, Snow!
When I began defining my PhD topic, snow on sea ice presented a particularly difficult problem. The most widely used climatology was still largely based on in-situ observations collected before the twenty-first century. Airborne measurements and modeling studies were beginning to suggest that this historical climatology no longer adequately represented contemporary Arctic conditions, and I wanted to investigate the problem using satellite observations.
I therefore set out to develop a satellite snow-depth retrieval method. It was not an easy problem. At the time, the main microwave-derived information available to us consisted of sea-ice concentration and the temperature near the snow–ice interface. That alone did not seem sufficient to retrieve snow depth reliably.
Things became much more interesting when my supervisor met a scientist from the Danish Meteorological Institute (DMI) and learned that DMI was producing sea-ice surface-temperature products from thermal-infrared observations. Unlike the microwave-derived temperature near the snow–ice interface, the infrared product represented the temperature at the snow surface. He brought the dataset back and suggested that it might be useful.
I spent many sleepless nights trying to work out what to do with these two temperatures.
One night, I was lying in bed under a blanket with the air conditioner running, thinking: I have temperatures at two different levels. What can I actually do with them?
Then came the moment that changed the direction of my PhD. The temperature difference between the inside and outside of a blanket becomes larger as the blanket becomes thicker. Snow on sea ice should behave in a similar way. Eureka!
Developing that idea, I found that the temperature difference between the snow surface and the snow–ice interface could provide information about the ratio of snow depth to sea-ice thickness, assuming that the temperature at the ice–ocean interface remains close to the seawater freezing point. That ratio could then be combined with freeboard, the portion of floating sea ice above the water, to estimate snow depth and sea-ice thickness simultaneously.
That was the origin of what became our simultaneous-estimation framework (Shi et al., 2020). Later, we extended the framework so that snow depth, sea-ice thickness, bulk density, and freeboard could be estimated together (Shi et al., 2023).
A Connection with Denmark
A Connection with Denmark
While developing the simultaneous-estimation method, I had the opportunity to spend time as a visiting researcher at the Danish Meteorological Institute. This was made possible through an international networking program between Denmark and Korea for which Gorm Dybkjær had secured funding.
During that visit I also met Rasmus Tonboe, who has since become one of my closest scientific collaborators. I remain very grateful for that opportunity. In retrospect, the networking program achieved exactly what such programs are supposed to accomplish: years later I joined DMI myself, and Rasmus and I continue to work closely together. Our collaboration has extended beyond passive-microwave studies to improving the interpretation and retrieval of sea-ice properties from radar altimetry (Shi et al., 2024; Shi and Tonboe, 2025).
Finally, a Snow-Depth Record!
Finally, a Snow-Depth Record!
Satellite radar-altimeter observations were not available with sufficient temporal and spatial coverage to extend our simultaneous-estimation approach over the full period we wanted to study. We therefore began asking whether freeboard (or ultimately snow depth) could be inferred using microwave observations alone.
Our earlier work on volume scattering provided a way forward. Because microwave optical thickness depends on the physical propagation path through the medium, the scattering signature contains information related to the physical thickness of the scattering layer, which we could use as a proxy for freeboard. We used this relationship to infer freeboard from microwave scattering characteristics and then incorporated that information into the simultaneous-estimation framework.
The result was a 2003–2020 Arctic snow-depth record that revealed substantial departures from the historical climatology and contrasting regional trends across the Arctic (our 2021 study). Special thanks to Dr. Walt Meier and Prof. Al Gasiewski for their contributions to this work.
This body of work eventually became the core of my PhD.
Returning to the Atmosphere
Returning to the Atmosphere
After completing my PhD, I stayed at SNU for a year as a postdoctoral researcher. Much of that period was devoted to improving and extending my previous cryosphere work, but it also gave me an opportunity to return to an important part of my earlier scientific training.
I became involved again in atmospheric remote sensing, participating in projects on temperature and humidity profiling using hyperspectral infrared observations and on greenhouse-gas retrievals. In a sense, I was able to resume a path that had begun during my early graduate years but had been temporarily overtaken by my fascination with the Arctic.
I am particularly grateful to Prof. Sang-Woo Kim at SNU for giving me the opportunity to broaden my research in this direction.
Research Meets Operations
Research Meets Operations
I later moved to DMI to deepen my work in polar microwave remote sensing and sea-ice radiative transfer. One of the biggest differences I felt between university and a research institute was the responsibility for operational services. It involved tasks that were quite different from what I had been used to, and to be honest, I initially found it a little overwhelming.
At first, operational work could feel very different from the research I had known at university. But I gradually realized that operations do not replace research. At DMI, we conduct research while also taking responsibility for turning what we develop into reliable products that users can depend on. In our terminology, this transition is often described as Research to Operations (R2O).
One of the largest programs I work within is the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF). Among its sea-ice products, I am responsible for the OSI-404 sea-ice emissivity product series. Assimilating surface-sensitive microwave channels in numerical weather prediction requires reliable information about surface emissivity, but sea ice is particularly difficult because its emission depends on complex and highly variable snow and ice properties. OSI SAF therefore developed an emissivity retrieval and operational product to provide this information for numerical weather prediction.
My role involves both improving the underlying science and maintaining the operational service. Through this experience, I learned that an operational algorithm has to satisfy requirements that go far beyond demonstrating that an idea works. It must remain robust as sensors change, input data become unavailable, processing systems evolve, and users continue to depend on the output. I also came to appreciate the importance of thorough documentation for traceability and of continuous validation for quantifying retrieval uncertainty.
More broadly, I learned how research at scale often depends on close collaboration between universities and operational institutes. Exploratory scientific ideas can be combined with the long-term infrastructure, continuity, and user engagement needed to turn them into sustained services. I also found it rewarding to work on something whose impact is visible beyond the research community.
Becoming an Independent Researcher
Becoming an Independent Researcher
As I became responsible for shaping my own research, I also had to learn how to sustain it. That meant expanding my network, building consortia, and writing many proposals. Some were successful; many were not. Proposal writing requires a substantial amount of time and energy, and I gradually learned that rejection is simply part of the process. If I wanted to develop my own research directions, I had to keep trying.
One successful proposal became particularly important in shaping my own scientific direction. In 2024, I received a Post-Doctoral Overseas Research Grant from the National Research Foundation of Korea for a project titled “Algorithm Development Framework for Next-Generation Earth-Observing Satellite Missions Combining Machine-Learning and Physical Methods.” The idea was to build coupled observation operators, use them to generate synthetic satellite observations, and train data-driven models for the simultaneous estimation of atmospheric and sea-ice variables. The project allowed me to explore an idea of my own on a larger scale, and the preliminary work eventually motivated the PhD project that I later helped establish around coupled observation operators and data-driven simultaneous inversion.
Securing a project, however, is only the beginning. Beyond doing the science and producing results, I had to learn how to manage projects: planning resources, monitoring budgets, coordinating responsibilities, responding when things did not go as planned, and ultimately taking responsibility for delivery.
One important experience has been ESA's Copernicus Imaging Microwave Radiometer Level-2 Product and Algorithm Development (CIMR L2PAD) project. The international consortium, led by the Norwegian Meteorological Institute, is developing retrieval algorithms for ESA's future CIMR satellite mission. I am grateful to Thomas Lavergne for leading such a large consortium. Within it, I have had to define how DMI can contribute scientifically, plan how we execute our responsibilities, and find solutions when technical or organizational problems arise. This experience has taught me a great deal about contributing to, and taking responsibility within, a large international scientific team.
Student supervision has been another important part of becoming an independent researcher. Together with colleagues, I helped secure funding and initiate an ongoing PhD project developing coupled observation operators for passive microwave radiometers and data-driven simultaneous inversion. I also co-supervise another PhD student studying CO₂ variability over Denmark. His research is primarily model-based, and one of my roles has been to help broaden the work by incorporating satellite observations.
Through supervision, I have learned that being a good researcher is not necessarily the same as being a good supervisor. My students' scientific interests will not always be identical to mine, and they should not have to be. I need to listen to what interests them, understand where they want to develop, and help them find a path toward becoming independent researchers themselves.
My Vision
My Vision
Alongside these changes in responsibility, working more closely with operational services and application-oriented research has gradually broadened the way I think about satellite observations. In many applications, observation operators are designed around the state variables available from a particular numerical model. An observation operator is a forward model that translates geophysical states into satellite-observed quantities such as radiances or radar waveforms. This is practical, but it can also mean that the observable physics represented by the operator is constrained by the model's own description and simplifications of the Earth system. I have increasingly come to think that the problem should also be approached from the opposite direction: start from the physical variables and processes needed to represent the measurement faithfully, and then develop interfaces that connect this observation-physics framework to different numerical models and applications.
At the same time, I became interested in the transformation itself. An observation operator is usually regarded as a tool for another purpose: to retrieve a physical state, evaluate a model, or assimilate observations. But what happens to the physics when an Earth-system state is transformed into observation space? Which physical relationships remain observable, which become ambiguous or disappear, and how are the dynamics of the physical system expressed in the quantities that satellites actually measure?
This question has become particularly interesting as data-driven Earth-system science advances. Recent developments such as ECMWF’s GraphDOP/AI-DOP have demonstrated that useful predictions can be learned directly from histories of observations, without first constructing a conventional analysis or reanalysis state. To me, this raises a fascinating physical question: if prediction can take place directly in observation space, what does Earth-system physics look like there?
My longer-term vision is therefore to develop physics in observation space as a research direction, using physics-based observation operators not only as links between models and measurements, but also as tools for understanding how physical information and constraints are transformed into what we observe from space. I would like to identify which physical relationships remain identifiable after this transformation and explore whether these constraints can make observation-driven prediction more interpretable, robust, and physically grounded.
(Updated 18 August 2026)
Page updated
Report abuse