CDE researchers Asst Prof Gianmarco Mengaldo, PhD students Jiawen Wei and Chenyu Dong, and collaborators from Brown University, Centre for Climate Research Singapore, and NVIDIA AI Technology Centre are using deep learning models to investigate the causes of extreme weather.

Their ongoing collaboration has resulted in their publication, XAI4Extremes: An interpretable machine learning framework for understanding extreme-weather precursors under climate change, listed by AI World as one of the week’s top 10 notable research papers!

As extreme weather events become more frequent and intense, enhancing our ability to predict them is crucial.

The researchers are investigating these weather events, namely heatwaves, using interpretable machine learning, specifically post-hoc interpretability methods applied to a binary time series classification deep learning (DL) framework they proposed. This approach produces relevance maps highlighting what input data the DL framework deemed important for its prediction.

The team will compare these maps with existing knowledge to see if they reveal new patterns that help us better understand the causes (also referred to as precursors) of extreme weather. These weather maps will also be organised by different multi-year periods to explore how climate change affects these precursors.

It is heartening to see collaborations such as this across organisations and involving young upcoming talents in research addressing some of the world’s most urgent climate concerns.

Jiawen, the paper's first author, says, “I have worked in interpretable machine learning and explainable AI (XAI) for the past two years. I am excited to start seeing the results of my research for scientific applications, and I look forward to contributing further to the intersection of XAI and science.”