Please somebody at @nature.com just read this

doi.org/10.1029/2024...

There may have been some coastal events, I can’t remember. AFAIK the event models were only forced with gauged river discharge so have basically nothing to say about surge events.

They were modelled directly using river gauges. Seemed to do an ok job all things considered (published here: nhess.copernicus.org/articles/21/...), but the sample size is *way* too small (not to mention the quality too inadequate) to give property-level true negatives

If that is solely based on the ~100? flood events shoved through the model back in 2018 then… just lol

Depends on the context. It doesn't follow that new / 'good' models simulate larger flood extents than old / 'bad' ones.

Nothing public as far as I can tell – I was emailed some NaFRA2 docs

Right! General idea is we’ll have “enough” events with a SMILE or two, and let the ML do the work at making the event magnitudes useable. All very much a WIP though. But avoiding the smorgasbord of choices in a Heffernan & Tawn type approach (eg doi.org/10.1029/2023...) would be nice!

Lastly, I had some fun arguing we should chuck the bathtub out with the bathwater (but keep the baby). Keep your HANDs to yourself!

Paper on this out recently with Brett Sanders & Paul Bates: doi.org/10.1029/2024...

Will do my best to fend off any whataboutery in the replies.

Seb Moraga on using a diffusion model to downscale a SMILE (yes, I love the acronym and no, I won't explain it)

Stochastic event sets from downscaled climate models? Yes please. So long & fuck you to arbitrary event definitions in statistical models.

#SWOT really is the real deal. Steve Chuter here showing how we can have real rather than made-up bathymetry for unsurveyed rivers.

A part of me is sad when something moves from parameterization to resolution. See you later 2y conveyance assumption. You made calibrating flow biases a lot easier 😔

On a more serious note, Chris Lucas showed us how FathomDEM is even better than FABDEM thanks to using vision transformers rather than random forest regression. It's quantitatively the best global DEM out there ... hold onto your hats for the paper (if the reviewers agree).

Also had the pleasure of meeting a #FABDEM superfan. He claimed to be the no 1 fan of FABDEM. I asked him to prove it, and he said he would do literally anything & was prepared for it to get weird. James: I have some ideas - call me.

iopscience.iop.org/article/10.1...

I'll start with my fav: Yasuo Nihei (Tokyo University of Science) chucking a daihatsu down a flume to see what would happen. I love science!

They've really sharpened my view of what good-faith actors in the climate risk industrial complex should try to achieve: treat models as metaphors, be transparent about value judgements, present results with humility, and make decisions which are robust to uncertainty - not paralyzed by it.

Thank you! Oh, man- gutted I missed out on the spicy bathtub discourse. Can we do it all over again?