Machine Climate

Department of Data Science, IISER Pune

We study how AI/ML weather models perform at medium-range and subseasonal lead times — a few days to several weeks out — with a particular focus on India and South Asia, a region shaped by monsoon variability and extreme weather. While the models we develop and work with are global, we maintain a particular focus that they be high-performing over South Asia, and especially for precipitation and temperature extremes.

Our goal is to make these models:

• Trustworthy, interpretable, with well-calibrated uncertainty, and robust to adversarial perturbations.
• Resource efficient, usable without large-scale compute, and accessible to researchers across the Global South.
• Explainable, aimed at understanding what makes these architectures work for weather and climate data, rather than primarily adapting large-scale models built elsewhere.