WeatherNext 3
deepmind.google
deepmind.google
I guesstimate that it has less than 50% accuracy for my area
As a former Googler, I wouldn't at all be surprised if this is an integration that is "planned" -- but just not done yet.
And some good handful of people are planning to wring a promo out of work. "Implemented weather UI in Android that is 63% more accurate." ;)
I'm building a tool right now that uses an ensemble to determine wind gust likelihood, which is useful for safety critical work on construction sites and the like.
I've heard some commentary in the past that 5G interferes with data measurements heavily used for weather forecasts (something about satellite measurements?)
I remember being _very_ impressed like 15 years ago about how I would have _hourly_ weather forecasts (in particular around the rain) that seemed like magic! And then things... seemed to slowly get worse (at least in Tokyo)
a couple years ago I was chatting with a friend in Kyoto. They used to live in Tokyo and had made the _exactly_ aligned comment like: "I was used to the rain forecasts being not so accurate anymore. After moving to Kyoto they seemed to be quite good! And now they're also bad here. Am I cursed?"
I looked at some 5G rollout maps and you could see Kyoto rollout happened a bit around the time frame they were complaining about....
Anyways I think for most people (at least for myself) weather forecasting seems like this odd dark magic that can't work at all, but there was a window in which it _felt_ like it was super accurate. At least in my personal experience
The short version of the "spectrum" issue is that 5G is being allocated in bands very close to the microwave spectra where atmospheric water vapor emits. A bevy of public and commercial satellites in low-Earth orbit passively monitor these microwave spectra and produce extremely important information that is assimilated into numerical weather models.
The federal government sets limits on "out-of-band emissions" for operators emitting in the allocated 5G spectra. These emissions can bleed over into the microwave water vapor bands, creating noise that masks the natural presence of water vapor. The limits for this out-of-band emission is on the order of ~10-20 dB, and there's some work in the atmospheric science literature suggesting that this is enough to confound certain water vapor retrievals. That basically means we lose these observations that help constrain the forecast.
There isn't much indication that this is a serious issue in day-to-day meteorology at the moment. But it's an issue which will be significantly more difficult to unroll and claw back than to simply protect key spectra in the first place.
On recent visits to Tokyo, with my current habits, the general crowd (who presumably check some app) has been the more reliable forecast. Either everyone has an umbrella in their hand or they don't. And I discover whether I should have brought my umbrella or if I'll be lugging it around pointlessly, just far enough from my hotel to be stuck with my choice
>The Google Weather forecast is created from an internal forecasting system that utilizes weather models and observations from global weather agencies.
It also lists the data sources it uses, but is vague about what model(s) it feeds the source data into.
So yeah, when it rains, it might take a few hours for that to flow into your weather app.
It's a direct result of DOGE.
Other apps have a different government?
I pull weather data from multiple apps all the time because I’m a weather nerd and they all agree equally poorly.
The accuracy of weather data has been largely unaffected since DOGE took hold. The reliability of the infra, however, has suffered somewhat. There are more outages now than there ever were, with a massive spike in outages, not kidding, the month DOGE was at its peak. At that time we even saw significant manifest changes - stupid stuff like clearly LLM-generated json fields that were whole sentences instead of the key/value pairs from the schema. It was a wild time.
Anyway downstream we're mostly fine nowadays. I can still follow storm tracks accurately, see real time responses, correlate with pilot reports and lightning strikes, compare with aircraft telemetry mostly favorably, etc etc. It's quite remarkable what high-end weather products are able to do, the ones you have to work for.
We do, however, merge several free/paid sources into one picture. Rarely do I find serious disagreement, but it does happen.
I ask alexa and it generally does fine. If uncertain I go outside for a bit and look at clouds.
For the details, you just have to know how your local weather works - thunderstorms come in squall lines here, or are hot-day flash storms.
Blizzards follow warm still weather the same way, but mostly in Jan-Mar, when the jet stream curls away and will snap back to bring arctic air down.
Moisture (rain/slush potential) follows California rain by 2 days or so. If my friends in cali complain about rain, I know we're gonna have school closures if in the winter.
If I really need to drill in on rain/snow chance, I google for a weather radar and do prediction in my own head. (windy.com is ok, accuweather is real data to watch). Storms generally move linearly or rotate, and gentle rain is wide while torrential rain is clustered.
Locally, many of the old wives tales about the shape of clouds or direction of wind vs prevailing (with against, cross-grain) are true enough for 12h-24h predictions in upper midwest USA.
If you use Google 99% of the time and only check other apps when Google is wrong, then that's a biased experiments.
Google does update its forecast more frequently, and usually is as accurate as the others for the hour ahead. And the longer range 3+ day forecasts aren't noticeably worse either. It's just the intermediate 4 - 48 hour stuff that they're surprisingly terrible with, but that's exactly the time period I most care about in everyday use!
It really does come down to data source quality and access…that’s the crux of it.
* https://apnews.com/article/weather-forecasts-worsen-doge-tru...
* https://www.independent.co.uk/news/world/americas/us-politic...
A good book on the history of forecasting, The Weather Machine: A Journey Inside the Forecast:
Would that be practical for weather forecasting or not really?
What are the conditions at 5000 feet, 10000, etc? What is the location of the jet stream and its strength? The reduced number of (e.g.) weather balloons is hindering forecasts (per the links).
The data that would be most valuable to initial conditions is upper-atmosphere winds -- this is the kind of data given by weather balloons. In clear air there's no great way to measure this from either the ground or from space.
One important supplemental data source here are aviation reports, from planes flying at altitude and particularly trans-oceanic routes. When air traffic was largely curtailed during the early phase of the Covid pandemic, weather forecasting suffered a bit for the lack of data (see eg https://www.ecmwf.int/en/about/media-centre/news/2020/drop-a...).
https://www.dwd.de/DE/wetter/thema_des_tages/2026/9/6.html (German only)
There's also https://nickleenders.github.io/verisky-scoreboard/history.ht... which tracks the trends over time.
It features a couple of AI and NWP model forecasts for comparison.
Link?
edit: The link for the demo is as thus, https://deepmind.google.com/science/weatherlab
Here's a URL for the demo for those who -- you know -- like to click on links and see stuff happen: https://deepmind.google.com/science/weatherlab
404. That’s an error.
The requested URL was not found on this server. That’s all we know.
WeatherNext 3 is being integrated into the Google products and tools that billions of people rely on – like Search, Maps, and Gemini. The model is also available for enterprise use, across a range of different applications.
Gain access to high-resolution forecasts, without model setup. Including real-time operational data and historical forecasts.
Try WeatherNext 3 in BigQuery, Earth Engine, Google Maps Platform and Google Cloud Storage.
---
None of those words are links or demos. It's just noise.
(I'm not trying to be pedantic, but if someone is having trouble finding the button, the exact text is helpful.)
Also, here's where the button takes you: https://deepmind.google.com/science/weatherlab
https://i.imgur.com/IVv4y0n.png
Keep in mind both button are "the demo". One is for trying out the API yourself, the other is for seeing a pre-made dashboard.
It is built into Apple Weather after Apple purchased them.
Did Apple lobotomize the tech when they integrated it, or lose access to whatever upstream data Dark Sky had access to? Apple Weather is comically bad (though not nearly as bad as Google's weather searches).
I used to be able to go on walks without an umbrella because of how reliable it was at predicting when rain was going to restart. If I do that with Apple Weather, I get home wet.
https://www.theverge.com/tech/883089/acme-weather-forecast-a...
"We create a 3-channel image where the red channel represents velocity in the x-direction, the blue channel represents velocity in the y-direction, and the green channel represents change in storm intensity" and then use computer vision techniques to predict what happens next
post was written after a discussion here on hn: https://news.ycombinator.com/item?id=3187326
'Nowcasting' is an area of active research, both with machine learning and with physics-informed or visual flow approaches.
Part of the problem from the machine learning side is that these are _huge_ problems. NVidia's StormCast (https://research.nvidia.com/publication/2024-08_kilometer-sc...) works globally at kilometer scales, and you can imagine how big those grids are. Even with patch training, you're dealing with very large datasets.
At the same time, this is not exactly a high-profile area of research. National weather centres focus on actionable medium-range weather predictions, and meteorologists can look at radar themselves and perform mark-one-eyeball predictions for very short-range watches and warnings. Some private-sector actors will pay for short range predictions, but they're often looking for something hyperlocalized (e.g. weather at this particular construction site, for crane safety) or specialized (near-real-time cloud and wind predictions for renewable energy).
Most public-accessible weather predictions are downstream of either a public-sector effort (which doesn't internalize benefits, leading to under-resourcing) or a byproduct of another private-sector offering.
- Time is UTC rather than local by default.
- Units are imperial rather than metric (i.e. not basing it on users locale)
- The concept of init time is not initially clear at all, so it's confusing when you click the calendar icon to see the weather forecast for a future date only to see it isn't an option, you have to use the slider.
- When changing the sidebar to the "detailed" view, most of the hovering element is cut off by the container so you can't actually read it.
And it took longer to write this comment than it did to find these issues. It's fine since it's clearly marked as experimental, but I disagree that it's "really easy to use" :p
Everything seems dark and desaturated, as if the the only clue we have (color!) for matching things up has been deliberately reduced.
And then: It sure does feel like the legend has even more of whatever-that-is going on than the map does.
I find it difficult to look at the map and understand the information it relays at the same time.
Init time makes sense to me, it's when the forecast/model is initialized, letting you look back in the past to watch the change to the present and future.
The sidebar thing is weird, but it also can expand out.
This seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value.
I imagine a lot of predictive sciences are ultimately about mixing together a bunch of inputs to attempt to decipher some output. Do we end up being able to take stuff from here and figure out some new ideas about modelling the climate as a whole?
In a high-level view, it's the result of specialized decoding heads.
Traditionally one would take gridded forecast outputs, then process those with comprehensible actions like "find all local pressure minima in the ocean, then filter to ones which correspond to warm cores, etc." to infer (diagnose) the presence of a cyclone.
One problem with this is that gridded forecasts suffer from known biases and tradeoffs. For example, a forecast on a ~25km grid is just on the edge of being able to represent the eye of a hurricane (50km scales), and it certainly can't accurately represent the sharp transition of wind in the eyewall. That means that the forecast winds are almost certainly a smoothed (and therefore less intense) version of what observers would see.
The WN2 approach (paper: https://www.nature.com/articles/s41586-026-10953-2) adds a direct readout head to the model: given latent-space access to the full forecast, it tries to predict the bona-fide cyclone observations (https://www.ncei.noaa.gov/products/international-best-track-...).
It's kind of like a post-processing or bias correction (see for example https://www.ecmwf.int/en/about/media-centre/aifs-blog/2026/a..., which applies in physical space), but by having access to the model latent space and by being included in model training it is (probably!) higher-quality than a pure, after-the-fact approach.
> WeatherNext 3 will power weather features in Google Search, the Gemini app and Google Maps, as well as the Google Maps Weather API and Google Earth Engine.
https://dataconomy.com/2026/09/04/weathernext-3-ai-forecasts...
So I assume the main way would be googling "weather Los Angeles" and it will be powered by the WeatherNext 3 models
They also open sourced the last one and are doing B2B/enterprise arrangements so maybe other weather apps are experimenting with it.
Problem is that the WN3 grid is still quite rough (5km) - but a that's a brutal improvement for many places compared to many other global models.
Quite a few country-scale models go down to a 1-2km grid nowadays. This is very helpful in complex geography like mountains and alleys.
If model members were available, with all the usual measures, thats a fantastic place to start looking at serious inclusion. It doesnt seem thats the case, however.
That's a pretty apples-and-oranges comparison. One would almost always use a high-resolution regional model if you needed certain details for different forecasting applications like renewable energy.
It's also worth noting that the 5km outputs are from a model decoder head that was trained against temperature and dewpoint at surface stations. According to the Rasp et al (2026) preprint, this head was designed for continuous sampling; the choice of a 5km grid is arbitrary. What we don't actually know is how well the model handles shocks like a frontal passage or impacts from things like outflow from storms - or even evaporative cooling from precipitation. We are limited to the output that DeepMind publishes; we can't run the model and stress test these things on our own.
That's all a long way to say that the 5km resolution is (a) limited to temperature fields, and (b) we don't know if the "additional" resolution has any impact whatsoever on the phenomena that one would typically use a mesoscale-resolving forecast for.
Never let a good crisis go to waste for marketing, especially if the product you're promoting is contributing to exacerbating it.
TLDR: the input data for the model now also includes real-time observations (satellite and weather stations) on top of the typical (re)analysis data, improving model resolution, run frequency and timestep frequency.