Comment by cryptoz
3 years ago
Yes, but, significant work was done at an academic level between 2011 and 2017, researchers at universities, grad student projects, etc, that resulted in usable data being extracted. I worked on this problem for ~10 years off and on. It is absolutely doable.
Especially since you don't actually _need_ raw pressure data to be useful. Pressure rate of change over time is still useful and produces improved accuracy in some local forecasts.
But not only that, the researchers at UW under Cliff Mass found that you can do on-device ML to clean the quality of the data, remove errors, and live-adjust to MSLP on-device, without even needing the dense network of sensors nearby for error correction.
It's 100% doable, but it just takes some hard science and effort.
> That makes it extremely noisy for detecting slow moving metrics like the weather
Yes, but, the noise problem was solved half a decade ago.
Huh. This is why I love HN. I didn’t know about that but looking up these papers to see if this approach would be useful to some other stuff I’m working on with similar-ish noise problems that conventional algorithms aren’t working for.
Best of luck! Here's one paper I referenced for a starting point, may be the most useful one (not sure..been a while..) https://journals.ametsoc.org/view/journals/atot/35/3/jtech-d...