Kara Sulia, director of the xCITE Lab at UAlbany's Atmospheric Sciences Research Center, explains how artificial intelligence can help us understand the weather and risks it poses — and why human scientists are critical to ensuring the computers aren't learning the wrong patterns in the high-stakes field of weather prediction.
For many people drawn to careers in atmospheric science, their curiosity is kindled by a formative weather event during their youth — a blizzard or flood that leaves such an impression that they spend their adult lives chasing the mysteries in the clouds and wind.
But for Kara, it was as much about the math.
“I knew I liked math. I knew I liked science. And I thought, ‘Meteorology seems cool,’” she reflected on her decision to pursue it as an undergraduate major. “I thought I was going to be a forecaster, but then I realized I didn’t like forecasting. But I always really liked the fundamental math, the calculus.”
As she pursued her PhD in meteorology, Kara also quickly recognized the centrality of computer science to the work she hoped to do.
“The coding and the computer programming was always my most favorite part of graduate school and anything I did in undergrad — writing software or writing programs. When I got my job here at UAlbany, I knew I wanted to spend more time learning the best ways to write code.
I still understand the fundamental physics, but I also just loved the programming part of it. Because I wanted to do all those things and develop software and take all this interesting data and do something with it, I started taking computer science classes. This past May, I actually earned my bachelor’s degree in computer science from UAlbany and, right before our conversation, I just came from my first master's-level class. So I’ve really been developing a robust background in computer science and seeing how I can use computer science as a tool — and AI as a tool —to enhance my research and the research of those in my center. I really like the mixture of the two, and I also realize that a lot of my peers, especially when I was an undergrad, and now a lot of students, kind of struggle with the computer science component because it wasn’t built into the curriculum. We weren’t taught coding. But literally every single thing we do, every project we have — unless you’re out there counting raindrops on leaves — you are doing some level of computer programming. And it’s not just our field. It’s any scientific field or beyond.”
Kara is uniquely suited to run ASRC’s AI/machine learning lab because she understands the fundamentals of both the atmospheric and computer science at work.
“You don’t want the computer model making predictions that aren’t grounded in reality,” she said. “You want the actual predictions to be tied to the laws of physics.”
Learn more about Kara’s research interests and UAlbany's Atmospheric Sciences Research Center
Go inside UAlbany’s xCITE Lab
Read how Carly Sutter, a graduate student in Kara’s lab, used machine learning to analyze road conditions based on traffic camera images
Explore ASRC’s history on top of Whiteface Mountain
And watch a daily time-lapse video from ASRC's perch on the roof of New York
Explore everything happening on campus with the University at Albany Events Calendar
Research and interview by Mike Nolan
Headlines by Erin Frick
Audio editing and production by Scott Freedman
Photos by Patrick Dodson
Written and hosted by Jordan Carleo-Evangelist
0:01 Jordan Carleo-Evangelist
Welcome to The Short Version, the UAlbany podcast that tackles big ideas, big questions, and big news in less time than it takes to cross the Academic Podium. I'm Jordan Carleo-Evangelist in UAlbany's Office of Communications and Marketing.
Atop one of the tallest peaks in the Adirondacks, UAlbany's Atmospheric Sciences Research Center studies the clouds from the roof of the Empire State.
That's not even really a metaphor. Three stories above the mountain's 4,867-foot summit, the roof of ASRC's Whiteface Mountain Field Station is the highest roof in New York. From it, ASRC scientists have spent five decades studying problems like acid rain, climate change, and — more recently — the impact of wildfire smoke on our state.
It can be a harsh environment, and it feels like a different planet from the geothermal comfort of UAlbany's ETEC building about 145 miles south. That’s where ARC's associate director, Kara Sulia, surrounded by the hum of computers, not howling winds, oversees the xCITE Lab.
xCITE stands for Extreme Collaboration, Innovation and Technology. You don't need to remember that. But you should know that it means that Kara and her colleagues are working with powerful computers to help translate fundamental science done in places like Whiteface into tools that make sense to the people who need them. People like emergency managers in charge of road safety or private companies working in drone flight or renewable energy.
ASRC is a great example of the two faces of modern science. There's experimentation and observation, and then there's the increasingly powerful technology like AI that helps us make sense of what our instruments see.
Artificial intelligence is fueled by data, and thanks to our centuries-long obsession with the weather around us, weather data is a bounty for the machine. Combining it with other data sets on demographics, flood maps or pollution — helping us see patterns where they might save lives — is where AI shines.
In this episode, Kara explains why even as technology helps us fill in the gaps in what we can observe, we still need researchers on cloudy mountaintops doing the fundamental science.
We spoke with Kara about where AI can help us understand what has and maybe will happen with the weather, where it falls short and how for her, it started with the love of calculus and the physics that make the atmosphere flow.
2:36 Kara Sulia
When I was really young, I knew I liked math, I knew I liked science, and I was like, ‘Okay, meteorology seems cool.’ I always really liked the calculus, the fundamental science, how things work, all of that, and so I went into research. And part of my research was the coding. The coding, the computer programming was always my most favorite part of graduate school and anything I did in undergrad was writing software or writing programs. And when I got my job here at UAlbany, I knew I wanted to spend more time learning the best ways to write code. I wanted to improve the efficiency of my model. I wanted to make sure my model was faster, I wanted to make it more accurate. I wanted to be able to run a model twice as fast. Because I wanted to do all of those things and develop software and take all this really interesting data and do something with it, having the computer science component tied in with the atmospheric science was always important to me, and I always enjoyed it. And because I was such an advocate for it, my boss saw the potential and saw that I was the right fit to lead the xCITE Lab.
The xCITE Lab is ASRC's AI/machine learning laboratory. We specialize in software development, data analytics, anything that has to do with weather-related data. A lot of the research that's done at ASRC is fundamental atmospheric science — development of models, forecast generation, collecting the data — like through the New York State Mesonet. But there's a big gap between the actual fundamental research and data collection and the end user — so, the state, the private industry, general public. And so the xCITE Lab kind of steps in and helps with that component where we take that fundamental research data and we convert it to a value-added product where maybe we take the output of a really complex forecast and we generate some more probabilistic statistics, or we create a dashboard on which people can view the data, play with the data, interact with it.
Emergency managers might handle the data and might find it useful for whatever their end resource is. So we’re converting the data into a usable format so it can be used by the end user. My graduate student, Carly Sutter, in collaboration with the Department of Transportation, developed a type of machine learning model that does image processing to detect that road surface condition. So now instead of the plow drivers or the field techs or someone maybe looking at the 2,400 camera images, we now have a model that can automate the prediction and label the entire state according to what those cameras are showing.
Weather is big data and the data comes from both observations and models. You have observations like satellite, radar, surface detection like the New York State Mesonet is a good example of that. We use the Mesonet data in just about every single project that we have. It is incredibly useful because now we're about a decade worth of data, so now you're kind of encroaching on climate scale. You can start to do climatological analysis of the Mesonet data, so the impact, it's going to be even larger. Having that data across the state is incredibly useful. We're the only network I think that has cameras at our sites, and the cameras, they're one of the most useful sensors for AI because there's so many things that can be detected in the camera images. And then you can confirm what you're seeing in the images from the rain gauge, from all the other sensors that are at the site. On top of that, it's a really high-quality data set, and we finally have the computational resources within the last half decade or so to pull down that data, process it and do what you want with it.
As the fundamental atmospheric scientist, I am very cautious to use AI for weather because a lot of the AI methods sort of rely on the inputs being independent of each other, whereas in weather everything is so interconnected physically. You increase the temperature and it affects the pressure, but those might both be independent inputs into an AI model. Really good ways that you can use AI and machine learning for weather is we have all these models and they're massive and they're running across either the continental United States or across the entire globe. Because they're so large and they're run so frequently, the resolution can be not optimal. The highest we see is three kilometers. So for places like New York City, they want to have information down to the street level, and so AI could be really good at that sort of thing because that’s not based in physics, that's more a statistical technique, so it really helps with that.
Also, the speeding up the process. If you're running a forecast and you want to run it every hour, it might take 15 minutes or longer to run the forecast itself, and so now you're 15 minutes into the hour. Whereas if you train an AI model to predict the same way that a forecast model does, it's completed almost instantaneously. So it speeds up the way in which we can receive the forecast.
You can't necessarily replace the existing forecast with AI because you need that physics to come in and you can't just create the physics out of thin air. That's kind of a pun. So you need that input data. It's that you're not necessarily increasing the accuracy of the forecast, but you're receiving the information quicker, maybe at a higher resolution and adding layers on top of that. OK, I have my output, now I'm going to add Census data. Now I'm going to add data about the local structures and how it's going to impact the streets or the flood zones or all of that. That's where it's really useful. You're bringing in disparate data sets.
I do not want to see AI being run to generate forecasts without the input being actual data. There are many who say, ‘OK, once we have it trained up, we won't need the actual observations or the model data to continue to allow it to run. The AI will just make the forecast itself.’ And while it may do a good job now, there are so many deviations, and a lot of them are unpredictable in atmospheric science. And it's not going to be able to capture them. It's identifying patterns, but it's really out to lunch on the physics. I'm sure we've all been using ChatGPT, and how many times does it take me asking the same question in multiple different ways to get the answer that I'm looking for?
There needs to be some sort of human intervention and critical thinking when it comes to using and analyzing the output from these models. In addition to that, a lot of the things that we're seeing with these models — and what the models are trained to do —is they're trained to predict the most likely outcome, but especially with weather, you want it to predict the most unlikely outcome as well. Because that's when the biggest costs to humanity, the biggest costs in dollars and lives, that's where that comes in is the outlier events. It would have never predicted to Superstorm Sandy because it had never happened before.
I think the big value for AI in our field is improving the speed, improving the resolution. Maybe we can come up with better ways to do forecasting, but I think it's really, again, connecting to the end user and providing that gap fill between the fundamental science and how can we deliver to the decision-makers, the general public, the information that they need and in a way that they can digest.
10:10 Jordan Carleo-Evangelist
That was Kara Sulia, associate director of the Atmospheric Sciences Research Center at UAlbany.
Kara explained how AI can help us make sense of the weather and what worries her about the things that the machines may not yet understand about the complex interrelated factors that make for a stormy day.
To learn more about why Kara recently returned to the classroom to earn her bachelor's degree in computer science, and to see a live feed from the roof of New York at ASRC's Whiteface Mountain silo, be sure to check out The Longer Version in our show notes.
Before we go this week, my colleague Erin Frick will catch you up on the headlines from around campus.
10:47 Erin Frick
A recent analysis by Stanford University ranked 55 current and emeritus UAlbany faculty members among the top 2 percent of researchers worldwide across 22 scientific fields. The rankings factor in a researcher's career output as well as their scholarly productivity in a single year. Distinguished Professor Aiguo Dai, who studies Arctic climate in the Department of Atmospheric & Environmental Sciences, ranked no. 3 in meteorology and atmospheric science.
For the second year in a row, UAlbany was among 30 institutions included in the Princeton Review Mental Health Services Honor Roll. The honor roll identifies institutions that demonstrate a strong commitment to student mental health and well-being based on surveys of administrators and students.
Marina Petrukhina, SUNY distinguished professor and UAlbany's Carla Rizzo Delray Professor of Chemistry, has been named the recipient of the 2026 George A. Olah Award in hydrocarbon or petroleum chemistry by the American Chemical Society. The prestigious award, which is named after a Nobel Prize winner, recognizes outstanding research achievements in hydrocarbon or petroleum chemistry. Petrukhina’s work on nanographenes has important implications for the performance and safety of lithium-ion batteries.
Looking ahead to the final week of the semester…
Join President Rodriguez on Tuesday, Dec. 9 for a Community Coffee Hour and to announce the winner of the annual Student Holiday Greeting Card Contest. In the spirit of the season, you're encouraged to bring donations for the Northern Rivers holiday giving campaign. New, unwrapped toys, clothing, home goods, and hygiene products will be collected to benefit children and families across the Capital Region. The event runs from 9-10 a.m. in the Campus Center Great Hall.
On Wednesday, Dec. 10, there will be a Build a Budget Workshop from 2-4 p.m. in Campus Center Room 367, where attendees can learn how to construct a budget to manage their personal finances, including understanding loan interest. This workshop is sponsored by the new Thrive at UAlbany initiative, which emphasizes health, career readiness, financial wellness and belonging.
If you're still on campus for finals, take a study break to watch the women's basketball team take on the Boston University Terriers at the Broadview Center from 2-4 p.m. on Saturday, Dec. 13.
You'll find links to all these stories and more in the Today at UAlbany News Center and a link to the full University Events Calendar in our show notes.
13:05 Jordan Carleo-Evangelist
The Short Version would not be possible without contributions from many people, including this week…
Scott Freedman, who provided audio production and editing support from the UAlbany Digital Media studio, deep inside the podium tunnels. Research by my colleague Mike Nolan, who — although you didn't hear his voice — interviewed Kara for this episode and covers UAlbany's weather and climate experts. And a news update from Erin Frick.
We hope you enjoyed this pilot season of The Short Version, and we'll be back in mid-January with more weekly episodes on spooky art, the origins of money and what the heck fruit fly brains can teach us about the way humans see the world.
Have a happy and safe new year.
I'm Jordan Carleo-Evangelist here at the University at Albany, and this has been The Short Version.