[bmgxyz]

This seems to imply that there isn't any high-resolution precipitation data available that could provide these "minute-by-minute" forecasts, but that isn't true. The National Weather Service provides several radar products that give data with resolutions in the range of 500 m using their NEXRAD technology[0]. This allows for some pretty good estimates of when precipitation will start and end over the next hour or so. This kind of forecast product is called a precipitation nowcast. Other nations have similar systems.

If you use the NOAA desktop tool[1] to view the data from NEXRAD stations, you can compare to services like DarkSky and see that they are very likely using it without much editing.

The simplest nowcasts use optical flow techniques rather than meteorological modelling. On short time scales (less than an hour), these methods can give passable results. I built a tool[2] that pulls this NWS data from their Web server and gives you a nowcast.

[0]: https://en.wikipedia.org/wiki/Nexrad#Super_resolution

[1]: https://www.ncdc.noaa.gov/wct/

[2]: https://github.com/bmgxyz/threecast

[srl]

Moreover, I've used minute-by-minute forecasts myself, and (at least where I lived at the time) they were quite accurate. My use case was "where is a 30 minute gap when I can walk home without getting soaked", and I never once got soaked. So much of the OP falls into the category of things-I-know-to-be-false-from-experience.

The article really has a strong vibe of "algorithms are faulty, we need humans in the loop to make sure they're behaving well!", with a hidden assumption of "humans are less faulty than algorithms". That's an empirical assertion to be determined on a domain-by-domain basis. It's certainly true that having a human in the loop leads to worse outcomes in chess (unless the human has enough modesty to just not do anything). The same is increasingly true of other domains as well.

Perhaps someday, incorrect, largely content-free FUD articles about how algorithms suck will themselves be written entirely by algorithms.

This is pushing way too many of my buttons, so I'll just close by pointing out (on the other side of the apps/humans scale) that a substantial fraction of the time, when I check the weather on NWS, it says something like "Today's high: 56; current temperature: 58". I certainly hope that a human in the loop would fix that problem.

[CrazyStat]

> It's certainly true that having a human in the loop leads to worse outcomes in chess (unless the human has enough modesty to just not do anything).

No, this is actually totally false. There is a world championship in computer-aided correspondence chess [1], and you won't get anywhere near the top ranks by having "enough modesty to just not do anything."

[1] https://www.iccf.com/

[cehrlich]

I can’t comment on America, but when I lived in Amsterdam everyone I knew used an app called Buienradar for this exact use case. The accuracy was astonishing.

[polishdude20]

Any chance of this working in Canada? Even just right across the border?

[bmgxyz]

I hadn't tested it before, but yes, it looks like there's coverage for some major Canadian cities near the border, including Montreal, Ottawa, Toronto, and Vancouver. You can check your locations of interest on this map[0] (requires JS). Just click "Maximum Radar Ranges" on the left and see if you fall within the coverage circles.

As I write this, there's some light precipitation just north of Toronto, and threecast seems to give reasonable output there. Note that the radar coverage areas aren't always perfect circles because terrain can block the radar beam.

Canada may have a similar system with better coverage, but I'm not sure.

[0]: https://www.ncei.noaa.gov/maps/radar/