Written by Dr Craig Hansen, New Zealand’s leading AI educator, bicultural digital specialist, and founder of the Summit Institute.
More than 1,000 schools across New Zealand trust Craig and the Summit Institute to navigate the complexities of generative AI strategy, policy, and whole-school roadmaps.
—
When storms strike island nations with little warning, the difference between an hour’s notice and six hours can be the difference between life and death. Google’s latest AI model is changing that calculus — and it doesn’t need a supercomputer to do it.
The Problem: A World Unequally Warned
Every day, the weather influences billions of decisions. Some are as simple as grabbing an umbrella. But across the Pacific — from Fiji to Vanuatu, from Samoa to the Cook Islands — weather decisions are anything but simple. Cyclones, storm surges, and flash flooding arrive with frightening speed, and traditional forecasting has long struggled to keep pace in regions where the infrastructure for high-resolution prediction simply doesn’t exist.
As Google’s own team notes in their announcement of WeatherNext 3:
“This breakthrough is particularly vital for regions across Latin America, Africa, and Asia-Pacific that have historically been underserved by high-resolution forecasting due to the immense supercomputing costs of traditional regional models. It brings localized, high-fidelity forecasting to billions of people and local businesses in these areas.”
That sentence carries enormous weight for the Pacific. Traditional numerical weather prediction (NWP) models — the physics-based simulations that have powered weather forecasting for decades — require supercomputers costing millions to run. For nations where GDP per capita is measured in the thousands, not the tens of thousands, that infrastructure has always been out of reach. The result? Communities in the path of a cyclone learn about it hours after their counterparts in Wellington, Sydney, or Honolulu.
Enter WeatherNext 3: Forecasting Without the Supercomputer
On 3 September 2026, Google DeepMind and Google Research introduced WeatherNext 3, described as “the most advanced and accurate global weather model to date, according to independent live evaluations by Brightband.”
The model’s arXiv paper, authored by Stephan Rasp and 24 colleagues, outlines why it represents a fundamental shift rather than an incremental upgrade:
“State-of-the-art AI weather models have shown impressive medium-range forecast skill and computational efficiency, but suffer two key shortcomings: their forecasts have lower spatial and temporal resolution than the best physics-based models and they are exclusively initialized with and trained on analysis data.”
WeatherNext 3 solves both. Here’s what changes for storm-prone regions:
1. Hourly Updates Instead of Six-Hour Lags
Traditional NWP models — and even previous AI models like WeatherNext 2 — produce forecasts every six hours. That six-hour gap matters enormously when a tropical cyclone is intensifying. WeatherNext 3 ingests live geostationary satellite data and generates a new forecast every hour, grounded in the most recent observations available.
As the blog post explains:
“By ingesting a mosaic of live, global geostationary satellite data, our new model gains a rich, continuously updating view of the atmosphere. This allows the model to generate a new forecast every hour, each one grounded in the most recent satellite observations available, at up to 5-kilometer resolution.”
For a Pacific Island nation watching a depression form to the northeast, hourly updates mean tracking rapid intensification in near-real-time — not waiting half a day to learn the storm has become a cyclone.
2. Five Times Sharper Resolution
WeatherNext 3 visualises key surface variables — temperature and moisture — at a 5-kilometre resolution, with other surface variables at 10 kilometres and atmospheric variables at 25 kilometres. The previous model operated on a 25-kilometre grid in 6-hour increments.
“Overall, this provides a global weather picture roughly five times sharper than our previous model, WeatherNext 2.”
For islands that are themselves less than 25 kilometres wide, the difference is not academic. A 25-kilometre grid might place an entire island in a single pixel, erasing the distinction between a storm hitting the windward coast and the leeward shore. At 5 kilometres, that same island resolves into meaningful terrain — and the forecast with it.
3. Real Observations, Not Inherited Biases
The arXiv abstract highlights a critical technical breakthrough:
“WeatherNext 3 moves beyond traditional analysis variables by learning to predict satellite-derived precipitation estimates, as well as tropical cyclone and station observations. Modelling sparse station data allows WeatherNext 3 to make 2m temperature and dewpoint predictions at any location and time, conditioned on local geographical features, with substantially lower error than competing global models.”
Previous AI models trained on NWP output — meaning they inherited every bias those physics simulations carried, including the six-hour data lag that can misrepresent fast-changing variables like rain or surface temperature. By learning directly from real-time satellite observations and sparse weather station data, WeatherNext 3 bypasses those inherited limitations entirely.
Saving Lives: The Practical Stakes
The Google team is candid about why hourly updates and sharper resolution matter:
“This is important because critical weather develops fast. When storms, fronts, or precipitation systems materialize suddenly, our rapid update cycle and higher resolution provides earlier, more detailed insights needed to help drive an effective response.”
For Pacific Island nations, “an effective response” is not an abstract phrase. It means evacuating coastal villages before a storm surge hits. It means fishermen deciding whether to stay in port. It means emergency managers pre-positioning supplies. The World Meteorological Organization has documented repeatedly that early warning systems reduce disaster mortality — and the gap in forecasting quality between the Global North and the Pacific has always been an early warning gap too.
Cyclone Winston (2016) killed 44 people in Fiji. Cyclone Pam (2015) devastated Vanuatu. Cyclone Harold (2020) tore through the Solomon Islands, Vanuatu, Fiji, and Tonga. In each case, the warning window — measured in hours — was the single most important variable in determining survival.
WeatherNext 3 does not eliminate the risk. But it narrows the information gap between Suva and Sydney, between Port Vila and Perth.
The Cost Revolution: Why Accessibility Matters
Here is perhaps the most underapprecated dimension of this announcement. The arXiv paper notes that WeatherNext 3 achieves its results with “impressive medium-range forecast skill and computational efficiency” — a phrase that understates a dramatic economic shift.
Traditional regional weather models for the Pacific require supercomputing infrastructure that costs millions. WeatherNext 3 runs on Google Cloud and is accessible through Google Search, Gemini, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine — many of them free to end users. The blog confirms:
“We’re making global weather predictions, updated hourly and ready to integrate into your workflows with no model setup required. This enables researchers, developers and businesses to query the data in BigQuery and Earth Engine, or bulk-download from Google Cloud Storage.”
No model setup. No supercomputer. No six-figure licensing contract. A meteorological office in Apia or Port Moresby can query hourly, 5-kilometre-resolution forecast data the same way a startup in San Francisco does. The democratisation of forecasting infrastructure is, quietly, one of the most consequential equity stories in AI this year.
Precipitation: The Hardest Problem, Now Sharper
Rain and snow are notoriously difficult to predict — so much so that the Google team devoted an entire section of their announcement to it:
“Global weather models notoriously struggle to accurately predict precipitation. Rain and snow systems are driven by fast-moving cloud processes on tiny scales that are hard to model accurately using traditional physics-based simulations. Consequently, AI forecasts often produce blurry estimates or miss the boundaries of severe storms entirely.”
WeatherNext 3 addresses this by training on NASA’s satellite-based IMERG data and Google’s own precipitation reanalysis. The result? A Continuous Ranked Probability Score (CRPS) improvement of up to 60% against IMERG for medium-range forecasts. For communities where flash flooding is the primary cyclone-related killer, sharper precipitation boundaries are not a technical footnote — they are a evacuation-order trigger.
Clean Energy: Powering Resilience After the Storm
Beyond disaster preparedness, WeatherNext 3 introduces forecasts specifically engineered for renewable energy production:
“The model forecasts 100-meter wind speeds (roughly at turbine-height) for precise wind-energy output, alongside high-resolution cloud cover and sun radiation levels to help solar farms estimate how much light they will receive on the ground.”
For Pacific Island nations pursuing energy independence through solar and wind microgrids — critical for resilience when cyclones knock out diesel-dependent power systems — these forecasts offer a new planning tool. Knowing how much wind a turbine will generate, or how much solar radiation will reach a panel array, helps island grids balance storage, demand, and generation across the cyclone season.
Education: Building Weather Literacy on Open Data
The accessibility of WeatherNext 3 through Google Earth Engine and BigQuery opens doors that extend well beyond meteorology. For educators across the Pacific and beyond, the model’s open data creates new possibilities:
- Classroom weather science: Students in Pacific Island schools can access the same hourly, 5-kilometre forecast data that powers Google Search — turning abstract lessons about cyclone formation into real-time, locally relevant observation. When a depression forms north of Fiji, students can track it through the same data their emergency managers see.
- Climate education: The integration with Google Earth Engine means teachers can overlay historical weather patterns, precipitation trends, and cyclone tracks on interactive maps — giving students a visceral understanding of how climate change is intensifying storms in their own region.
- Data science curricula: University programmes across the Pacific can use BigQuery-accessible WeatherNext 3 datasets to teach data analysis, probability, and environmental science using real-world data from their own neighbourhoods rather than imported examples from the Northern Hemisphere.
- Community preparedness training: Adult education and community outreach programmes can build weather literacy using the same model that informs official advisories — closing the gap between technical forecasting and community understanding that has historically left Pacific populations underprepared.
Google’s stated goal reinforces this educational potential:
“Our primary goal is to advance weather intelligence to make it universally useful — whether for an emergency responder tracking sudden wind shifts, an air traffic controller planning flight paths, or a farmer managing crops.”
“Universally useful” includes the classroom, the community hall, and the village evacuation plan.
The Honest Caveat
Google’s announcement closes with a disclaimer worth repeating:
“For official weather forecasts, severe weather warnings, and public safety advisories, please refer to your local meteorological agency or national weather service.”
WeatherNext 3 is a tool, not a replacement for institutional forecasting infrastructure. National meteorological services remain the authoritative source for warnings and advisories. But WeatherNext 3 narrows the data gap that has made those services harder to run in under-resourced regions — and gives them a sharper, faster, cheaper foundation to build on.
What This Means for the Pacific
For communities that have lived for generations at the mercy of storms they could see coming but not forecast in detail, WeatherNext 3 represents something quietly profound: the same quality of weather intelligence that wealthy nations have long taken for granted, now available without the supercomputer price tag.
Hourly updates. Five-kilometre resolution. Direct satellite learning. Free or low-cost access through Google Cloud and Earth Engine. No model setup required.
The atmosphere will always retain a degree of unpredictability — Google’s team says so themselves. But for the first time, the gap between what Wellington knows about an approaching cyclone and what Nuku’alofa knows about the same cyclone is measured in minutes, not hours. And in the business of saving lives, minutes are everything.
Sources
- Google Blog: “Introducing WeatherNext 3, our most advanced and accurate global weather AI model.” The WeatherNext team, Google DeepMind & Google Research. 3 September 2026. https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/
- arXiv Paper: Rasp, S., Babenko, B., Masters, D., et al. (2026). “WeatherNext 3: Increasing resolution and performance of global weather models with raw observations.” arXiv:2609.03582 cs.LG]. [https://arxiv.org/abs/2609.03582
- Independent evaluation: Brightband live forecasting leaderboards — https://owb.brightband.com/
- Developer resources: WeatherNext 3 API and data access — https://developers.google.com/weathernext
🎓 Partner with New Zealand’s Leading AI Educator
Are you a principal, school board member, DP, or teacher leader in primary, intermediate, secondary, or tertiary education looking for practical, secure, and culturally grounded AI integration? Dr Craig Hansen and the Summit Institute provide bespoke professional development (PD), custom AI Policy workshops, or school-wide strategic implementation planning explicitly designed for the New Zealand educational context.
Contact us today to reclaim your teacher workload bottlenecks and establish secure learning guardrails:
👉 Explore AI Training & Workshops at summitinstitute.ac.nz | 🎬 Watch the Full Policy Video on YouTube Channel (@summitinstitute)
