Dr. Gan Zhang
As an assistant professor of climate, meteorology, and atmospheric sciences, and leader of the Climate Modeling, Prediction, and Applications research group (C-MPA), Dr. Gan Zhang is dedicated to investigating weather extremes and their impacts using a wide range of datasets and machine learning tools. In this article, Zhang shares his experiences and thoughts on the state of weather and climate science, two closely connected fields at a pivotal moment of challenge and promise.
Key Takeaways:
- Data Quality and Collaboration — Effective forecasting depends on high-quality data collected by U.S. and international partners, combined with global data-sharing efforts.
- Synergy of AI and Physical Models— Advanced storm-resolving physical models and AI innovations are poised to facilitate collaboration and increase prediction accuracy and accessibility.
- Private Sector Growth — While the migration of weather and climate scientists from the public sector to the private sector seeds innovation, privatization risks create service gaps for less profitable communities.
- Funding Shifts Pose Challenges — Shifts of federal support may threaten operational and research capabilities, raising concerns about public safety. “Weather and climate fields in the U.S. are experiencing huge technology and funding disruptions,” Zhang said.
- Resilience & Optimism — Zhang highlights the dedication of public sector workers and believes new collaboration between physical and computational researchers offers a path toward a more resilient future.
The world of weather and climate science is undergoing a significant transformation, driven by the rapid advancement of artificial intelligence and set against a challenging federal funding environment. In this story, we’ll talk about the changing landscape and how weather and climate data are aggregated, shared, and used.
Dr. Gan Zhang is not only an expert on weather extremes and their impacts, but also knows them firsthand. His childhood near the Huai River in East China, a region marked by distinct monsoon seasons and prone to flooding, inspired his career in weather and climate science.
He began his studies at the Ocean University of China, and completed his master’s and PhD in Atmospheric Sciences at the University of Illinois. After graduating in 2018, he moved to the East Coast and worked as a postdoctoral researcher at Princeton University. He was affiliated with and stationed at a federal lab that pioneered weather-climate modeling. In 2020, he joined the hedge fund, Citadel, as a weather analyst serving commodity trading and eventually returned to the U. of I. as a professor and leader of a research group.
Zhang’s work mainly focuses on physical and data-driven modeling, which can serve stakeholders in the public and private sectors. For instance, probabilistic predictions and risk assessment of high-impact extremes are increasingly used to inform risk management and investment decisions. At the razor’s edge of this science is the ever-improving quality of model technology, and now the combination of artificial intelligence with traditional physical models.
High Resolution Models to Unify Weather and Climate Modeling
Much of the most destructive behavior of the atmosphere occurs at scales of a few hundred extending down to approximately one kilometer—a spectrum that low-resolution physical models struggle to faithfully capture. However, Zhang highlights that the development of new global storm-resolving models helps represent such intricacies and can ultimately serve as a digital twin of the Earth.
“[With low-resolution models], you tend to miss lots of details, such as hurricanes, severe thunderstorms, and other weather extremes,” Zhang said. “That has been an issue for decades.”
Zhang explained that this new development has led to a large volume of simulation data that strains the storage of supercomputers. Since the late 2010s, such simulations have gone from covering just 40 days–a period too short for analyzing climate statistics– to more than a decade of simulation length. The sheer data volume inspired global collaboration exploring the best practices of data sharing and workforce training.
“In the past, weather and climate scientists went in different directions because we had been prioritizing working on different physical processes, but now, because of these very high-resolution simulations, we will have more common ground.”
In May 2025, Zhang revisited Princeton to attend the World Climate Research Programme Global KM-scale Hackathon, which happens at ten nodes around the globe. The activities aimed to promote global collaboration, enhance accessibility to high-quality data resources, and expand long-term access to the data-sharing environment.
Thanks to such collaboration, specifically between modeling centers in the U.S., Europe, Asia, and other regions, students and scientists interested in the new high-resolution simulations got hands-on experience. They generated numerous analytics that offered scientific insights and helped model developers.
“So suddenly many attendees recognize that we can use these storm-resolving models to do climate study,” Zhang said.
Because of this level of quality and promotion of accessibility, it has provided equity in research and a space for experts to talk closely and engage in each other’s research, keeping everyone on the same page.
“In the past, weather and climate scientists went in different directions because we had been prioritizing working on different physical processes, but now, because of these very high-resolution simulations, we will have more common ground,” Zhang said.
“(Local data) are helpful for one or two-day forecasts in the US. But, if you are thinking about two weeks, several months ahead, lots of information comes from other regions of the globe.”
Data Sourcing and Butterfly Effect
To unlock the potential of these high-resolution models in life-saving operational prediction, researchers and engineers need to feed their models with vast quantities of high-quality observational data—from satellite observations to direct measurements collected by aircraft flying into hurricanes.
“For physical prediction models, if you feed them with poor initial conditions, the forecast will not be so great,” Zhang said. “The measurements are key for unlocking the potential of these high-resolution models.”
Zhang emphasizes that observational data has to be strategically collected – sometimes from the eye of the storm.
“You need to collect the best data from key regions. Not all the data from the world matters the same for a two-day US hurricane forecast,” Zhang said. “NOAA and other agencies have been flying planes into hurricanes to collect such data. These in-situ real-time measurements are a key contributor for the performance of high-resolution models.”
For longer forecasts stretching weeks or months ahead, however, a global perspective built on international collaboration and data sharing is essential.
“(Local data) are helpful for one or two-day forecasts in the US,” Zhang said. “But if you are thinking about two weeks, several months ahead, lots of information comes from other regions of the globe.”
Weather researchers have long recognized that forecasts can be sensitive to distant, small-scale flow perturbations. The mathematician and meteorologist Edward Lorenz famously described this as the butterfly effect. The poetic term was coined in 1972, when Lorenz noted that the details of a Texas tornado could be influenced by minor perturbations such as a butterfly flapping its wings in Brazil.
Since the 2010s, climate researchers have increasingly recognized such sensitivities in climate simulations. To quantify their impacts on climate trajectories, climate researchers run dozens, sometimes hundreds, of parallel simulations starting with slightly different initial conditions. Some emphasize the number of parallel simulations, while others invest more computation power in model resolution to better assess high-impact extremes.
Zhang and his collaborators have been using these large-ensemble, high-resolution climate simulations to quantify hurricane risks.
“These include slow-moving storms like Hurricane Harvey, which dumped 40 inch rainfall in Texas… [These models] can also better represent hurricane-induced inland rainfall, such as what we saw with Hurricane Helene,” Zhang said. “The model simulations that we examined suggest such events get worse if strong warming occurs.”
Atmospheric modeling usually ends where the rain meets the land.
“But that’s where societal impacts start. Hydrologists use the rainfall predictions from weather and climate models to inform decisions ranging from life-saving evacuation to real estate investments. More accurate, granular predictions enabled by high-resolution models and global observations are invaluable,” Zhang said.
The Rise of AI Models and New Contributors
As physical scientists continue improving their models, big tech companies are stepping in as new disruptive forces, bringing artificial intelligence into weather and climate modeling. Zhang described how these innovations were initially met with skepticism, then unprecedented enthusiasm.
“In January (2025), we had this big annual conference of the American Meteorological Society. Everyone was talking about the AI models’ remarkable performance and potential,” Zhang said. “These sessions were always packed…people were standing in the hallway trying to get a peek into the future.”
For now, Zhang describes AI models as an exciting aid to traditional physical research, not a replacement.
“You have a lot to gain by unifying the vision from the two sources of information,” Zhang said. “The most exciting current development (for weather forecasting) is really a marriage between the best physical-AI models and the best observation system. Besides national agencies and big techs, start-ups are also very active in the space now.”
For climate modeling, one of the most exciting developments is the increased accessibility of simulations, says Zhang.
After working with new AI models, Zhang proposed the upgrade of a high-performance computing cluster, “Keeling”, at the University of Illinois. Using new GPUs acquired in early 2024, Zhang and his team executed simulations of extremes at a scale once unimaginable outside major federal institutions.
“The bottom line today is that I can set up a pipeline within half a day and make a decent prediction of weather in Champaign or elsewhere. That’s running a single GPU for several minutes or less,” Zhang said. “We also had remarkable success in using the same AI models to make seasonal predictions of hurricane activity.”
This marks a major shift from the past, when only national labs or resourceful institutions could afford the technical expertise and computational power to perform such tasks.
“In the old days, when I was working for the NOAA lab, the first step of research was to simulate hurricane activity with very (resource-)heavy physical models. These models required thousands of CPUs to run,” Zhang said.
Because of AI, a power once reserved for a few institutions has opened up, allowing for unprecedented innovation.
“So in a way, with the AI models, more developers and users can run their own simulations for innovative use cases. It still requires technical expertise, but at least now many more have the capability to start the job. The improvements of AI models are unbelievable,” Zhang said.
Still, AI models are not without limitations, and experts are not without hesitations.
For more consequential events like extreme weather, AI models are being carefully evaluated in the process of operational prediction. Such efforts include a newly announced collaboration between the National Hurricane Center and Google.
“Hurricanes have real impacts on people’s lives,” Zhang said. “We want to make sure we get the best prediction possible. The AI models haven’t been fully verified to see the potential strengths and weaknesses in every aspect.”
The fast adoption of the technology was hard to imagine just a few years ago.
“Initially, I would tell everyone, [AI] is the future for weather prediction… two or three years ago. Some nodded politely, but no one really believed in my conviction,” Zhang explained. “Then one day my father, who is a retired engineer, asked me if I knew AI could make great weather predictions…The rest is history. I would say the field and the public are having an open mind, and the conversation has shifted a lot in the last two or three years.”
“When it comes to life-saving services for communities with different characteristics, we should make sure changes are thoughtfully introduced.”
Innovation and Shifts in the Wake of Funding Cuts
Like most federal agencies, NOAA and other environmental science institutions have experienced or face looming budget cuts, threatening a drought of valuable data.
“NOAA is the biggest player in the world. The US has an extensive observation network, mostly operated by NOAA,” Zhang said. “It is hard to estimate the true scale impact, but we do hear anecdotal recounts that the operation of some observation stations were threatened or affected by staff cuts or other issues.”
NOAA, which manages the operational flights into hurricanes to take measurements that aid forecasts, was forced to cut hurricane-operation staff this spring, according to Zhang. However, without having entered the peak of hurricane season yet, it is hard to say how the cuts may strain the system.
“Maybe it appears fine to have fewer staff before the hurricane season, but when things get really busy, say three hurricanes active in the Atlantic basin, you might start to have problems,” Zhang said. “The situation is still fluid. It is very hard to estimate the potential impacts at this point.”
Budget cuts have cost jobs and prompted many in the weather and climate fields to pivot toward technology, finance, and insurance, coinciding with a large recruitment effort by these industries for their skills.
Zhang made a similar transition at the peak of the COVID-19 pandemic from public-sector research to a hedge fund. Reflecting on his experience and his peers’, Zhang shed light on the impacts of weather and climate scientists leaving academia for the private sector.
“Many great people take their scientific insights and rigorous skills to the private sector,” Zhang said, “these new contributors, working with their new colleagues, can push the service for business customers to the next level. At a personal level, working in the private sector needs adaptation and can be challenging. But some, including me, find such environments stimulating.”
The talent migration also highlighted the possibility of deep privatization. A concern Zhang has is whether essential services for the less populated or less resourceful areas become underprioritized because of the incentive toward profits.
“If the weather and climate services are completely privatized, business organizations would naturally focus on the markets that can bring in the most profits,” Zhang warned. “New York City, and other major population centers, get more customers, lots of coverage, and lots of attention, but smaller rural communities like those in the Appalachian Mountains may receive less attention, much less likely to get high-quality service by the enterprise.”
This is where you need a government agency like NOAA to hold the essential service for communities, according to Zhang.
“When prototyping AI tools, I can ‘move fast and break things’. Unhappy users can raise an issue at GitHub to get fixes or have a rollback,” Zhang said. “But when it comes to life-saving services for communities with different characteristics, we should make sure changes are thoughtfully introduced.”
“This is an extraordinary community. They have taught me many lessons about innovation, resilience, and commitment.”
History Notes and Long-Range Outlook
Historically, weather forecasts concern how to predict the future based on precise knowledge of the current state, and climate predictions depend on slow-changing forces that govern the overall system. The future of weather and climate science in the U.S. has more uncertainty than at any other time. Zhang acknowledges ongoing shifts and welcomes more hands on deck, and anticipates new players to bring in more technology and advancements.
“During my transition back to academia, I went through the early history (of my field) and life stories of founders. One of them is John von Neumann, who is known as ‘the father of the modern computer’ and did many fascinating things. One of his first projects on the first modern computer was weather prediction. He also drafted the proposal to establish the NOAA lab in Princeton, where I worked,” Zhang recalled.
“I read his 1955 proposal for federal administrators and thought, ‘wow, von Neumann foresaw so many milestones in weather and climate modeling’. Of course, it’s hard for von Neumann to foresee the recent AI innovation. Even compared to five years ago, the field now is much broader, has a big audience, and lots more exciting technology,” Zhang said. “I truly see an opportunity for the field to evolve in a direction that better serves society.”
Zhang’s confidence in the rapidly evolving field comes from the people he’s met along the way.
“The computer science researchers are extremely smart, passionate, and collaborative… For the other side of the aisle, the transition and adaptation—both individually and institutionally—are challenging in many ways. But people in the field have a track record of accomplishing extraordinary things, even in the most adverse environments,” Zhang said.
His belief in their resilience is rooted in the people’s dedication. From flying into hurricanes to collect data to learning new AI technology, those people dedicate their careers to science research and improving life-saving predictions. With such commitment and resolve, Zhang is confident the field and its people can eventually get through volatile times.
“Most people, including me, flee when seeing a hurricane coming,” Zhang said. “But my colleagues choose to fly into hurricanes. This is an extraordinary community. They have taught me many lessons about innovation, resilience, and commitment.”
Learn More and Get Involved
Contact Dr. Gan Zhang here.
Contact the Office of Data Science Research if you’re aware of other people or resources we could feature here. ODSR is a campuswide convening organization that facilitates collaborations, resource sharing, and public engagement focused on data science research activities at the University of Illinois.