On February 25, VIEWS hosted a virtual Brownbag seminar featuring a presentation on the working paper “Rapid Machine Learnt Climate Emulators Enabling Broader Risk Assessments.” The paper explores the role of machine learning in enhancing climate risk assessments and was presented by Dr. Vassili Kitsios.
The research underscores the urgent need to assess climate risk under a wide range of economic and emissions scenarios, particularly as the global economy transitions toward net-zero emissions. While climate models under the international Coupled Model Inter-comparison Projects (CMIP) provide essential data for understanding future temperature and rainfall patterns, they are limited in the number of scenarios they simulate. To address this challenge, the study introduces QuickClim, a machine learning-based climate emulator that rapidly reconstructs gridded climate data for various emissions pathways.
QuickClim facilitates broader assessments of physical climate responses and downstream human impacts, with applications spanning agriculture, finance, health, and social unrest. The seminar covered both the development of climate reanalysis data assimilation systems and the application of data-driven methods to quantify climate impacts on human systems. The research was recently published in Nature Communications Earth and Environment and featured in a Nature News Feature alongside other international advancements in AI-driven climate modeling.
Learn more:
- “A machine learning approach to rapidly project climate responses under a multitude of net-zero emission pathways” (Kitsios et al., 2023)
- “How AI is Improving Climate Forecasts” (News feature in Nature by Wong, 2024)
- “CAFE60v1: A 60-Year Large Ensemble Climate Reanalysis. Part I: System Design, Model Configuration, and Data Assimilation” (O’Kane et al., 2021)
Illustration: Jasiek Krzysztofiak. Hurricane: Rawpixel; Globe/computer code: Pexels; Weather map: US Dept of Commerce, Weather Bureau (CC BY 2.0)