Hacker News Reader

Comment by uticus | Hacker News Reader

> So how did every reliable model we use miss this? That’s for graduate students and researchers to answer in the coming years, because I have no formal idea.

Would love to see followup on this. Wish there were some way to like bookmark or something and come back in a few years.

View on HN

I'm one of the researchers on a new disease forecasting project, and I'm bookmarking this because we often use hurricane forecasting as an example of a level of accuracy we aspire to, so this is a really interesting breakdown.

The big miss here seems to be that not enough data had been seen yet to even know the envelope of what's possible. Without that there will always be inputs where a model breaks down. Disease modeling is also always going to have to contend with the fact that our historical data was collected when things weren't quite what they are today, for instance population numbers and densities are still entering uncharted territory compared to the past.

If author's hunch was correct, it may have been a "greater than the sum of its parts" hyperlocal edge case.

E.g. the wind sheer exactly right, a patch of warm water, etc.

That does seem really tricky to model without making the models overly jumpy.

What forecasting models / techniques are prevalent in disease forecasting?

There's usually a mix of statistical models (some traditional time series methods, some neural network based) and mathematical models (i.e. SIR models). None of them fully capture all the dynamics we need to.

Differential equations make the world go round.

That, and conservation laws. (Angular momentum)

there is already a very good theory. see this interview with an expert from Yale: https://youtu.be/CSVFpval4Dw?si=oelutbqzOGcpI8gm

tldr: garbage data into the model, garbage out.

Detailed summary from interview: • only 1 hurricane hunter flight was done into this storm. • mexican hurricanes just don’t get as much reporting coverage (historically). For example for atlantic hurricanes, NOAA hurricane hunter airplanes have tail doppler radar, which provides good data to feed into the model. • without detailed doppler radar, they had to fallback to satellite measurements. • Mexico doesn’t have ground based radar coverage at all, like the US coast does.

This is the most likely answer. Since the atmosphere is a chaotic system weather models are extremely sensitive to initial conditions. A lot of the improvements in weather forecasts have come from better data assimilation for better initial conditions.

too late to edit my comment, but after more reading it seems like mexico does have land radar along the coast, but it’s a much smaller network compared to the US coast. So for this particular hurricane, there was lack of coverage.

Yep, I commented here that my company, Care Weather, is working on that problem. Goal is to provide hourly updates of ocean surface winds.

https://news.ycombinator.com/item?id=38038558

Does Acapulco airport not have radar because the ATC for that part of Mexico is at Mazatlán?

I meant ATC sector HQ, of course Acapulco has a tower.

i’m not an expert in this area myself. your comment made me curious though. chat gpt (GPT 4) helped me understand better:

https://chat.openai.com/share/9448215d-2b65-4611-a634-f1a900...

Those locations are nowhere near Acapulco.

I think that was part of the point of the person you're replying to.

Post it on Reddit and use the reminder bot.

https://www.reddit.com/r/RemindMeBot/comments/e1bko7/remindm...

Or put it on your calendar with an email alert ;)

add a calendar event for when you want to be reminded

As far as I know some models gave the storm a 25% chance of rapid intensification. The actual storm itself just unraveled afterwards.

>The actual storm itself just unraveled afterwards.

Isn't this the expected behavior once it made landfall?

It basically intensified a lot more than 25% I mean.

Yes, that's expected.

Reply on news.ycombinator.com