Machine learning for climate systems
2020-04-02 — 2025-10-02
Wherein satellite shortwave‑infrared retrievals are described and daily detection of methane super‑emitters of at least 10 tonnes per hour is shown, and ML‑driven control of chaotic weather systems is surveyed.
How we model the world with data-hungry methods. How we can think our way out of the climate crisis.
1 ML for climate simulation
-
Climate Informatics is an open community interested in research combining climate science with approaches from statistics, machine learning and data mining. Through the annual conference series and through this community website, we hope to stimulate discussion of new ideas, foster new collaborations, grow the community, and thus accelerate discovery across disciplinary boundaries.
2 ML for climate driver identification
ESA - Trio of Sentinel satellites map methane super-emitters
In a recent paper published in Remote Sensing of Environment (Schuit et al. 2023), researchers from SRON found that the Sentinel-3 satellites can retrieve methane enhancements from its shortwave infrared band measurements. Impressively, it can detect the largest methane leaks of at least 10 tonnes per hour, depending on factors like location and wind conditions, every single day.
See also Pandey et al. (2023).
3 ML for climate interventions
I just saw Miyoshi presenting about this Japan Moonshot program on weather control. See Coverage in IEEE Spectrum.
Miyoshi and Sun (2022), Sun, Miyoshi, and Richard (2023) are interesting proofs-of-concept showing we can achieve high-efficiency control of chaotic systems in simulation.
Their methods are not particularly “ML-like”; I’m curious how much better we could do.
The possibility of “switching off” typhoons before they make landfall is very interesting.
4 ML for climate solutions
Jeff Dean’s NeurIPS 2019 talk suggests ideas. His talk is basically an advertisement for tensorflow probability as a toolkit for machine learning for physics simulations aimed at making nuclear fusion feasible, etc.
5 Incoming
Yanir Seroussi, The mission matters: Moving to climate tech as a data scientist
Gencast/graphcast