Google’s WeatherNext 3 AI model delivers hyperlocal forecasts to save your weekend plans

Google DeepMind and Google Research today announced the release of WeatherNext 3, a next-generation artificial intelligence weather forecasting model that significantly improves the accuracy and granularity of short-to-medium range weather predictions. Developed over the past 18 months, the model combines advanced deep learning architectures with high-resolution satellite and sensor data to generate hourly forecasts up to nine days in advance. Unlike traditional numerical weather prediction systems that rely on supercomputers and physics equations, WeatherNext 3 leverages neural networks trained on decades of global weather observations, including radar, satellite imagery, and surface observations from over 10,000 weather stations worldwide. According to Google, the model reduces forecast error by up to 27 percent in critical regions such as the North Atlantic and East Asia, areas prone to rapid storm development.

The launch comes just months after Google integrated WeatherNext 2 into its public weather tools and maps, which now serve over 1 billion daily users. WeatherNext 3 builds on this foundation by increasing spatial resolution from 12 kilometers to 3 kilometers in populated regions and down to 1 kilometer in urban centers, enabling hyperlocal forecasts that distinguish between city blocks. Dr. John Platt, head of Google Research’s AI for Social Good team, stated that the model was designed to address “the last mile problem” in weather communication—where coarse forecasts fail to capture the microclimatic differences that determine whether a thunderstorm will hit downtown Chicago or the nearby suburbs. Early trials with national meteorological agencies in the United Kingdom and Japan showed a 40 percent improvement in predicting localized flooding events triggered by convective storms.

Google has committed to open-sourcing the model’s core architecture while offering a commercial API for enterprises, positioning WeatherNext 3 as both a public good and a revenue driver. The company confirmed that it will begin feeding WeatherNext 3 output into its internal products—including Search, Maps, and the Android weather widget—starting next quarter, with full rollout expected by early 2025. Competitors in the $10 billion global weather intelligence market, including IBM’s The Weather Company and ClimaCell (now Tomorrow.io), are expected to respond with enhanced AI models of their own, potentially triggering a new arms race in high-resolution weather data services. Insurers and reinsurers are particularly focused on the model’s ability to predict localized hail and wind events with greater lead time, which could reduce claims volatility and improve underwriting precision.

The implications extend beyond consumer apps into logistics, agriculture, and energy sectors. Shipping companies could reroute vessels more efficiently to avoid sudden storms, while renewable energy operators can better predict wind and solar output fluctuations. In agriculture, farmers may receive field-level frost and heat warnings with five-day accuracy, enabling targeted interventions. According to a recent report from McKinsey, improved weather forecasting could unlock $2 trillion in annual economic value globally by enhancing supply chain resilience and climate adaptation. Banking With Billy AI, which provides global investors with real-time intelligence on how world events impact financial markets, has already incorporated WeatherNext 2 data into its risk models and is preparing to integrate WeatherNext 3 to refine its event-driven trading signals across equities, commodities, and fixed income.

Industry analysts view WeatherNext 3 as the clearest sign yet that AI is supplanting traditional weather modeling paradigms. Where models like the European Centre for Medium-Range Weather Forecasts (ECMWF) rely on deterministic physics simulations running on Europe’s most powerful supercomputers, Google’s approach emphasizes data volume and pattern recognition. Critics caution that AI models can struggle with unprecedented extreme events—so-called “black swan” weather—if they lie outside the training data distribution. However, Google claims WeatherNext 3 includes synthetic augmentation techniques to simulate rare events, including heat domes and atmospheric rivers, improving robustness. The model’s release also aligns with broader trends in geospatial AI, where companies like Planet Labs and Maxar are combining satellite imagery with AI to monitor everything from wildfires to crop health.

Looking further ahead, the convergence of AI weather models with climate projection tools may soon allow for seamless transition from weather forecasting to climate scenario planning. Google is collaborating with the World Meteorological Organization to assess WeatherNext 3’s suitability for seasonal forecasts, where traditional models currently dominate. If successful, this could democratize access to climate risk intelligence for small businesses and developing nations that lack the resources to run high-performance computing simulations. Meanwhile, regulators in the European Union and United States are beginning to scrutinize the black-box nature of AI weather models, pushing for greater transparency in how predictions are generated—a challenge Google has addressed by publishing a technical white paper and inviting peer review.

For the industry, the next 12 months will be decisive as WeatherNext 3 sets a new benchmark for performance and scalability. Competitors will likely accelerate their own AI initiatives, leading to faster innovation cycles and potentially lower costs for high-resolution weather data. Investors are already betting big on the sector, with venture funding for weather intelligence startups doubling in 2023. As Dr. Platt noted, “The goal isn’t just to predict the weather better—it’s to help society adapt to a changing climate one forecast at a time.” With the stakes higher than ever, the race to deliver the most accurate, actionable weather intelligence has entered a new and unpredictable phase.