A few years ago, in the 2017-19 timeframe, android phones had the best "next few hours" weather prediction I've ever seen. It was way more accurate than wunderground, accuweather and all other web services. Sometime after 2019 it seems to have gone, and I wonder what happened.
Speculation: goog used the barometric sensors in many phones "near you" to increase the precision of their models, making "immediate timeframes" extremely precise.
No idea if this actually happened or it was confirmation bias on my part, would love for someone with knowledge to chime in. I also wonder why they stopped, if my speculation is correct. Data gathering stuff, perhaps?
> Speculation: goog used the barometric sensors in many phones "near you" to increase the precision of their models, making "immediate timeframes" extremely precise.
They didn't.
Smartphone pressure observations (SPO) have extremely limited use in real-world meteorology for a whole variety of reasons. First and foremost - to actually use them, you have to assimilate them into a numerical forecast model. Very few commercial organizations outside of specialty weather companies do this, and fewer still run their own assimilation systems. The most well-known claims about incorporating SPO data into an operational forecasting system come from The Weather Company, but there's limited information in the public domain about what they _actually_ do.
The problem is that we know there are big problems with SPO data. Cliff Mass had several PhD students in the late 2010's that looked at this in detail (e.g. [1] and [2] are good entrypoints to the body of literature this group produced). The best summary I can offer is that (1) SPO data requires on-device calibration and bias correction otherwise it's relatively unusable in downstream applications; and (2) even when you _do_ incorporate SPO data into high-resolution simulations, they have little to no impact on forecast skill or quality.
There has been some work recently that uses SPO data for post-hoc analysis of weather events (e.g. [3]; IIRC there is a nice Google Research blogpost about this too, but I can't find it immediately). But that's a very different application.
Google likely just worked with a vendor that had nowcasting capability (which was very in vogue due to the popularity of Dark Sky). But all those forecasts are literally just simple extrapolations of radar imagery, and are only useful for precipitation.
A few years ago, in the 2017-19 timeframe, android phones had the best "next few hours" weather prediction I've ever seen. It was way more accurate than wunderground, accuweather and all other web services. Sometime after 2019 it seems to have gone, and I wonder what happened.
Speculation: goog used the barometric sensors in many phones "near you" to increase the precision of their models, making "immediate timeframes" extremely precise.
No idea if this actually happened or it was confirmation bias on my part, would love for someone with knowledge to chime in. I also wonder why they stopped, if my speculation is correct. Data gathering stuff, perhaps?