This weekend, Therese and I took a walk, and because the weather forecast from the day before had said there might be rain, we brought an umbrella. But it didn’t, in fact, rain. And that got me thinking: forecasts generally get more accurate the shorter-term and more localized they are, so a two-hour forecast, at most an hour old, of the precise area we were in, could probably have told us that there was in fact little chance of rain.
It turns out there’s already a name for really-short-term forecasts: nowcasts.
People like air traffic controllers already get these kinds of forecasts, and depending on where you live, you can already get a forecast for the next couple of hours. But I haven't found any really fine-grained forecasts freely available online, and geographical resolution really does matter: for example, if you’re five kilometers further west, it might start raining half an hour later.
With the recent proliferation of phones with GPS and Internet connectivity, this could become a useful service: Wouldn’t it be immensely convenient to be able to glance at your phone, and be told that yes, those dark clouds are likely to get you wet within the next hour?
I guess the real question is whether the weather models we already have can produce this information—without much in the way of human input, since we’re talking about producing separate forecasts for every square kilometer of land.