When Climate Data Comes Alive: A Day at TACC

The roar of cooling systems in the Texas Advanced Computing Center (TACC) isn’t just noise—it’s the power of supercomputers translating the overwhelming dimensionality of climate data into something students can finally see and understand.

Author: Koutian Wu; GitHub: ktwu01

On April 10, 2026, our GEO 371/391T Climate Data class ventured to this facility, one of the largest university supercomputing centers in the U.S., for what Professor Geeta Persad described as a transformative experience. What started as a routine field trip became a moment of scientific clarity where abstract data transformed into tangible understanding.

The Challenge: Making Sense of Dimensional Chaos

Climate data isn’t just big—it’s bewilderingly multi-dimensional. Our task for the day was to explore extreme precipitation patterns using datasets that spanned:

• 2-D space (latitude and longitude across the globe)
• 1-D time (historical records through future projections)
• 4 observational datasets and 10 different climate models
• ~30 realizations of natural variability (because Earth’s climate is inherently chaotic)
• 3 future emission scenarios (optimistic, middle-of-the-road, and business-as-usual)

We focused specifically on rolling time series of 1-day, 3-day, and 5-day cumulative rainfall at the 95th percentile—essentially tracking extreme precipitation events.

Morning: Setting the Stage

The day began with an introduction to the challenge ahead. Professor Persad outlined our mission: explore extreme precipitation patterns across multiple dimensions of climate data. The scale of what we were about to tackle was daunting, but the energy in the room was palpable.

Francesca Samsel, a research scientist at TACC and trained artist, then introduced students to the art and science of data visualization. Her slides immediately engaged students, particularly when she demonstrated how strategically designed colorbars can reveal patterns that poor color choices obscure. Using examples from Matthew Larsen’s work at Lawrence Livermore National Laboratory, she showed how the same dataset could tell completely different stories depending on color scale choices. It wasn’t just about making things look pretty—it was about making data readable and scientifically meaningful.

The HPC Tour: Where Climate Science Gets Real

Before diving into data analysis, we toured TACC’s supercomputing facility. Walking into the machine room was like stepping into the engine room of a massive ship. The roar of cooling systems hit us immediately—a constant, thunderous hum that made conversation nearly impossible. “It’s like standing next to a jet engine,” one student shouted over the noise.

Rows upon rows of servers blinked in synchronized patterns, their LED indicators creating a mesmerizing light show. The cooling system was particularly fascinating—some racks used liquid cooling with visible tubes carrying coolant through the systems. Students learned that a single high-resolution climate model run for 100 simulated years can take weeks of real time on these machines and consume enough electricity to power a small town.

“Wait, so when we’re running our Jupyter notebooks on TACC, we’re using these?” a student asked, gesturing at the supercomputers. Yes—though our classroom exercises used a tiny fraction of the capacity. The scale suddenly made sense: you need this much computing power because you’re solving fluid dynamics equations for millions of atmospheric grid cells, thousands of times per simulated day, for decades or centuries of simulation.

The tour gave crucial context for what came next. This wasn’t just an abstract exercise—we were about to work with data generated by these very machines.

The First Struggle: Parallel Coordinates and Information Overload

Back in the visualization lab, we started with PyCinema, which uses parallel coordinate plots to help navigate multi-dimensional data. The initial reaction from most students? Confusion.

One group spent nearly 45 minutes just trying to understand what each axis represented. “Wait, so this line represents one model’s prediction for one location under one scenario?” asked a student, tracing a path through the tangled web of lines on screen. The answer was yes—and there were hundreds of such lines.

“I personally think we had the hardest one,” Sarim Ahmad, an undergraduate student, remarked during a moment of frustration, half-joking but also genuinely overwhelmed by the complexity. It was the kind of comment that captured the day’s challenge—this wasn’t meant to be easy.

The breakthrough came when students realized they could interactively filter the data. By selecting specific ranges on different axes, they could isolate patterns. One team investigating Southeast Asian monsoons discovered that when they filtered for high 5-day cumulative rainfall, only certain models showed agreement with observational data. “It’s like the models are arguing with each other,” one student remarked. That’s exactly what model uncertainty looks like.

The “Aha” Moment: From Spaghetti Plots to Scientific Questions

The real learning happened when groups started formulating their own research questions. Each team approached the data differently, wrestling with the complexity in their own way.

[Need to gather more thinking from students]

(KW Note: group sort is clockwise as we did in final pre)

Mock Group example asked: “Which climate models best match historical observations for extreme rainfall in the Amazon basin?” They discovered that the GFDL model showed strong agreement for 3-day events but diverged significantly for 1-day extremes. This led to a discussion about model resolution and convective parameterization—concepts that suddenly mattered because they explained what students were seeing.

Group 1 tackled xxx.

Group 2 explored xxx.

The Technical Hurdle: Paraview and 3D Visualization

After isolating interesting data subsets in PyCinema, students moved to Paraview for high-fidelity 3D visualization. This transition proved challenging, but it also showcased the collaborative spirit that made the day special.

“I can’t figure out how to make the continent boundaries visible on the globe,” admitted a student from Group 3, staring at a sphere rendered in rainbow colors but lacking geographic context. The technical barrier was real—Paraview’s interface isn’t intuitive for first-timers.

What happened next was exactly the kind of peer learning we hope for. Pranav Rajkumar, an undergraduate from Group 3, walked over to Group 2’s workstation to ask for help with the same mesh visibility issue. Through collaborative troubleshooting—comparing settings, sharing tips from the TACC documentation, trying different rendering options—they figured it out together.

Professor Persad, Cameron Cummins (who had driven down from Dallas to support the session), and the TACC science and visualization team provided guidance when students got stuck, but the actual problem-solving came from the students themselves. This kind of collaborative learning, where students teach each other through shared struggle, often sticks better than any lecture.

A New Appreciation for the Work

The hands-on experience gave students a visceral understanding of what climate scientists actually do. Raha Helali Esfahani, a master’s student who spent much of the session working on creating her own precipitation maps, captured this sentiment perfectly: “Maps are really difficult. I have so much more respect for the scientists who do this work.”

Her comment resonated throughout the room. The field trip transformed abstract appreciation into concrete understanding. What looked like simple colored maps in papers and presentations turned out to require wrestling with data formats, coordinate systems, color scales, and countless technical decisions. Each choice mattered.

Despite facing funding challenges in recent years—a reality many graduate students navigate—Raha remained hopeful and enthusiastic about pursuing PhD research in the US. She had discovered something important about herself: after teaching for three semesters back home, she realized that while teaching was interesting at first, the repetition wasn’t for her. “I don’t want to be a professor,” she explained candidly. “I want to do research.” The experience at TACC reinforced this clarity—the challenge of wrestling with data, solving technical problems, and discovering new patterns was exactly the kind of work she wanted to pursue.

The Data Availability Problem

One of the most frustrating—and educational—challenges came early in the analysis session. Several groups had been briefed about specific datasets or models they wanted to examine, only to discover that the data wasn’t actually available in the system. This led to confusion and ate into precious time.

“We were informed about a specific dataset we wanted to examine, but the data was not actually available,” explained Sarim Ahmad. His group spent valuable minutes troubleshooting before realizing they needed to pivot to a different dataset entirely. “An extra hour would have allowed us to complete everything,” Sarim reflected afterward. “We eventually figured out how to fix the issues at the end, but by then we were rushing.”

As TA, I noted that this frustration was actually an authentic research experience. This is exactly what happens in real climate science work: datasets have gaps, documentation is incomplete or outdated, and researchers must adapt on the fly. The time constraints made it more stressful than it would be in a real research setting, but the lesson was valuable. Learning to pivot when your planned approach doesn’t work is an essential research skill—one that’s hard to teach in a traditional classroom setting.

The Lightning Talks: Science Communication Under Pressure

The day concluded with 5-minute “lightning talks” where each group presented their question, workflow, visualization, and conclusions. This proved harder than expected.

Groups that had spent hours exploring data struggled to distill their findings. “We found that… well, it’s complicated,” began one presentation. Professor Persad gently pushed back: “What’s the one thing someone should remember from your work?” This forced students to identify their core finding and articulate it clearly.

Another group created a stunning Paraview visualization but initially couldn’t explain what it showed scientifically. “It looks cool, but what am I learning?” Greg from TACC asked. The students regrouped and added annotations, highlighting specific regions where their chosen model diverged from observations. The revised version told a clear story.

The strongest presentation came from Titus Li’s group, which kept it simple: one clear question about Texas precipitation, well-filtered PyCinema plots showing their selection process, one focused Paraview visualization with a carefully chosen color scale, and three bullet points of conclusions. “Natural variability dominates near-term Texas precipitation uncertainty” was their headline finding, backed by visual evidence. The clarity came from ruthlessly cutting everything that didn’t directly support their main point.

What Students Actually Learned

Beyond the technical skills with PyCinema and Paraview, students gained several deeper insights about the nature of the work. They quickly realized that models are tools, not absolute truth, and that understanding where different models agree or disagree is itself a form of scientific knowledge.

They also learned that visualization is fundamentally a form of analysis. The process of creating a meaningful visual representation explicitly forces you to understand your data, reinforcing that pretty pictures without scientific insight are just decoration. Furthermore, the lightning talks proved that communication is an entirely separate challenge; understanding something yourself and communicating it clearly to others are distinct skills, yet both are essential for science.

Ultimately, students saw firsthand that real research requires continuous adaptation. The data availability issues and time constraints they faced weren’t bugs in the field trip design—they were features. Learning to pivot when a planned approach inevitably breaks down is perhaps the most essential research skill of all.

The Lights in Their Eyes

Professor Persad’s observation about seeing “lights in students’ eyes” captured something important. This wasn’t about passive learning or memorizing facts. Students were actively struggling with real scientific challenges—the kind researchers face daily. The frustration when PyCinema plots made no sense, the excitement when patterns emerged, the satisfaction of creating a visualization that actually communicated something meaningful—these are the experiences that transform students from consumers of science to practitioners of it.

One student summed it up during the bus ride back: “I finally get why climate scientists are always talking about uncertainty. It’s not that they don’t know anything—it’s that they know exactly how much they don’t know.”

That’s the light Professor Persad saw: the moment when climate data stopped being abstract numbers and became a window into understanding our planet’s future.

Acknowledgments

Special thanks to Francesca Samsel and Greg Abram from TACC’s visualization team for their patient guidance throughout the day. Francesca, a research scientist and trained artist at TACC, brought her expertise in “sculpting” data through artistic principles of color and form. Greg, with over 40 years of experience in computer science and visualization, provided the technical expertise to help students turn their ideas into actual visualizations. Thanks also to Cameron Cummins, who recently joined the Department of Earth and Planetary Sciences as a Computational Geoscientist, for making the trip down from Dallas to support our session. And finally, thanks to Professor Geeta Persad for designing a field trip that challenged us to think like climate scientists rather than just learn about climate science.