Improving Climate Projections Through Bias Correction: Insights from the Paris 2025 Workshop
From May 26–28, 2025, The Climate Data Factory participated in the 3rd Workshop on Bias Correction in Climate Studies at Mines Paris – PSL. The event brought together leading climate scientists, statisticians, and data practitioners from around the world to focus on a central question: How can we better correct biases in climate simulations to improve their usability and reliability?
Why Bias Correction Matters
Climate models are essential tools for understanding future climate risks, but they are not perfect. They often misrepresent regional patterns, extremes, or variable dependencies due to physical simplifications or coarse resolution. Bias correction techniques help bridge this gap between raw simulations and observed climate data — a step critical to trustworthy climate impact modeling.
Key Highlights from the Workshop
We were exposed to a rich variety of methods and use cases, including:
- Univariate vs. multivariate bias correction, especially for hydrology and agriculture
- Correction strategies for extreme events, including temperature extremes and compound flooding
- Trend-preserving methods such as CDF-t and emerging generative AI approaches
- New ensemble techniques, like α-pooling and Graph Cut optimization, to combine multiple climate models
- Spatial consistency challenges tackled with vine copulas and hierarchical modeling
- Real-world applications across French overseas territories, South America, and Africa
Our founder, Harilaos Loukos, was a co-author in the BADJAM study on how different bias correction methods affect crop model outputs. This collaboration with the Joint Research Centre in Ispra and international partners showed that multivariate corrections generally lead to more accurate and consistent agricultural projections — reinforcing the importance of method selection based on end use.
Our Takeaways
Participating in this workshop confirmed what we see in our work every day: bias correction is not just a technical issue — it’s foundational. The better we correct for bias, the more useful our climate data becomes for adaptation, planning, and policy.
The sessions also gave us a fresh look at emerging statistical tools, especially the move toward ensemble-aware corrections and high-frequency (hourly) adjustments. These advances will help climate data better serve the energy, health, and insurance sectors — areas where we’re already seeing demand grow.
What’s Next?
We’re incorporating what we learned into our methodology updates and looking forward to collaborating with several of the researchers we met in Paris.
We thank the organizers, especially Mathieu Vrac and the ESTIMR team at LSCE with whom we have been collaborating for years, for hosting an inspiring workshop.
