Shiqi Fang is a postdoc research associate at North Carolina State University. He is perhaps best known for his pioneering work to improve the accuracy of large-scale climate model projections. He is the first author of a new paper that addresses the new challenge of providing high-quality forecasts for compound extreme events. These collective events have the potential to greatly benefit society at large. Fang’s pioneering research explores the combined effects of temperature and humidity, two key factors driving these weather extremes.
Our interdisciplinary research team, featuring collaborators from NC State and the University of Arkansas, has pioneered an advanced machine learning approach. They dubbed it Complete Density Correction with Normalizing Flows (CDC-NF). With this approach, they increase the accuracy of climate models. It zooms in particularly on the more complicated compound events that can vary drastically from global to local scales. These discoveries proceeded in the magazine Scientific Data DOI 10.1038/s41597-025-05478-8.
Advancements in Climate Modeling
Fang’s research represents an important breakthrough in climate science by addressing these fundamental challenges involved in projecting compound events. In fact, these events usually happen when two or more physical variables collide, resulting in catastrophic effects. For example, the recent increase in extreme heat is worsening the impacts of heatwaves — the increased danger they pose to communities, for example.
The CDC-NF technique uses state-of-the art machine learning algorithms to downscale climate model projections. More specifically, it recalibrates the probability distributions of climate variables. This ensures that predictions do more than just mirror out the mean. The predictions get the extremes right, too. Fang and his team made sure to test rigorously. They demonstrated that utilizing the CDC-NF method quadrupled the predictive accuracy of all five climate models. This development demonstrates the ability of machine learning to tackle decades-old obstacles in climate modeling.
Testing Across Scales
These tests, performed by the research team, marked the first time the CDC-NF method was rigorously tested on a global scale. They further focused attention to its intended use inside the continental United States. This two-pronged approach served to challenge their first attempt at methodology, pushing them to test the impact on different contexts and projects. By training on data from one geographic region and testing it on data from other regions, they discovered that the approach reliably improved model performance.
Fang reiterated that this improvement is important both for policy makers and researchers. They rely on precise climate predictions to decide the best course of action. Better predictions can help us all be better prepared when disaster strikes. It’s no coincidence that these events are occurring with increasing frequency due to climate change. Prohibitive as they are, accurately projecting these compound events is essential for scientific credibility, safety, and ultimately, survival. It is key to minimizing their societal harms.
Collaborative Efforts and Future Implications
The research was co-authored by notable experts in statistics, including Emily Hector and Brian Reich from NC State, along with Reetam Majumder from the University of Arkansas. Their joint work really highlights the interdisciplinary aspect of pushing the boundaries of climate science using advanced statistical methodologies and machine learning.
The ramifications of Fang’s work reach far beyond the ivory tower. With climate change posing a growing threat globally, accurate models are essential for disaster response strategies and resource allocation during extreme weather events. Climate models are increasingly adopting techniques such as CDC-NF. This powerful integration produces increasingly reliable forecasts to aid communities in effectively adapting to a changing reality.
