2. Additionally you don't need to train a model, since most of the math is pretty well defined. All you need to do is initialize with latest data and let it crank. Where most of the development happens is in learning more about how the physical planet it set up, then adjusting your model to incorporate that newly known aspect
3. The weather forecasting market is actually rather robust, especially for an industry largely propped up by government grants. This probably isn't a great candidate for 10x disruption here and may not pass the "toothbrush test" either.
[1] https://www.ncdc.noaa.gov/data-access/model-data/model-datas...
looks like just a market to me which they could dip into and make some cash, without a deeper sense of vision nor a real change and impact. (which is where I usually see google/alphabet striving for)
maybe you see something deeper into that which i'm missing the point.
I choose my weather services based on presentation, because they all basically rely on the forecasting from NOAA [in the states, at least and from what I can tell].
In my imagination of Google, they could slurp up all the public weather data and make it better and more relevant and more present than other websites.
But the incentives are probably not there and my imagination is probably out of touch with the difficulty of chaotic systems.
We have lots of cool ideas at work which basically start (and end) with "if we could just convince a N thousand people to install this app and agree to continually send us all their phone data then it would be really easy to..."