Peter Watson

Machine learning applications for weather and climate need greater focus on extremes - topic of perspective just accepted in ERL: iopscience.iop.org/article/10.108… @IOPPublishing (with nicer formatting at arxiv.org/abs/2207.07390) 1/

@RuiZhuRich @dougmcneall See nercgw4plus.ac.uk/projects-2025-….

2026 PhD opportunity: "Machine learning for understanding extreme weather in a changing climate", to develop methods to apply new ML/AI tools in trustworthy ways. At Bristol Uni in the GW4+ PhD program. With interdisciplinary team inc. @RuiZhuRich, @dougmcneall. Links below.

Happy to share Bobby Antonio's paper on applying generative machine learning to improve skill at forecasting rainfall in East Africa: doi.org/10.1029/2024MS… Particularly interesting aspects are improvements in the diurnal cycle and getting good performance in an extreme season.

For full project description and instructions on how to apply, search for "Bristol PhD in Physical Geography" to find our department's page - X doesn't accept the link. Deadline is Jan 31

PhD opportunity at Bristol: "Machine Learning for Understanding Extreme Weather in a Changing Climate". Please share with potential applicants. With @RuiZhuRich, @milankloewer

@drexlspivy56 I don't think they show that.

@PeterAGWatson do we know yet if it can do better than conventional models after including statistical postprocessing?

Very exciting work on probabilistic weather forecasting by DeepMind's GenCast: nature.com/articles/s4158…. But I think most of the points from this thread still apply.

Also nice to see they've made the model code and weights available.

Some positive development (beyond having probabilistic forecasts) is evaluation at greater extreme thresholds - up to 99.99th percentile of 12 hourly data, or ~1 in 14 year events. So getting into the extreme tail, but there's still more to look into there.

Also now on BlueSky: @peteragwatson.bsky.social

With a great team including @paul_d_bates, @oejwing, @climatehat and Sean Longfield at the Environment Agency. Thanks if you can share.

PhD opportunity to study climate drivers of future extreme flood risk, using recently developed large-ensemble simulation datasets, machine learning methods and state-of-the-art flood risk modelling: noc.ac.uk/gsnocs/project…. In Flood-CDT (flood-cdt.ac.uk).

Only just learned there's a way to turn off allowing xAI to train their Grok model on your tweets, which was enabled by default. Under Settings -> Privacy... -> Grok

Very happy to share Henry's excellent work applying machine learning to achieve km-scale downscaling of rainfall in climate simulations! It's far cheaper than running km-scale regional models, opening up applications to e.g. studying extreme events in large simulation ensembles.

With the new academic year soon to start, it seems like a good point to crowdsource resources with advice for starting PhD students - is there anything out there that you think would be particularly useful to share? Anything you'd have liked to know?

Article 3's account of how governments and corporations have undermined freedom of speech and managed to stop people speaking out is also highly alarming.

Very interesting series from Vox. The first article on how environmental organisations currently greenwash meat production is certainly instructive about how big corporations corral others to control media narratives.

20yrs ago Valdes moved to Bristol and our climate/biochemistry group (BRIDGE) was born. We celebrated our anniversary by bringing back many of our alumni. I love co-leading the group with @OceanFMonteiro - life is a lot of fun when you have such friendly and exciting colleagues❤️

Our climate group has a new website so you know exactly what we are going 😀 - very proud to have so many amazing and fun people working together. Check it out: climatebristol.org Co-led by @_RachelJames @EuniceLoClimate @PeterAGWatson

With a great team: @paul_d_bates, @_sallybrown77, @JamesPope10 and @oejwing

PhD opportunity to study climate drivers of extreme UK flood risk and understand their role in future changes. Includes potential to apply newly developed machine learning-based rainfall prediction methods: noc.ac.uk/gsnocs/project…. In #FloodCDT, deadline March 20.