Key Insights from Weather Prediction Models

So excited to share our new Nature paper on GenCast, an ML-based probabilistic weather forecasting model: https://lnkd.in/enzPUFbn It represents a substantial step forward in how we predict weather and assess the risk of extreme events. 🌪️ GenCast uses diffusion to generate multiple 15-day forecast trajectories for the atmosphere. It assigns more accurate probabilities to possible weather scenarios than the SoTA physics-based ensemble system from ECMWF, across a 2019 evaluation period. It’s vital that we ensure these new ML weather systems are safe and reliable. One thing I'm proud of is our range of evaluation experiments: per-grid-cell skill & calibration, spatial structure, renewable energy, extreme cold/heat/wind, and the paths of tropical cyclones (i.e. hurricanes). For example, we created a dataset of simulated wind power data at wind farm sites across the globe, and found that GenCast outperforms ENS by 10–20% up to 4 days ahead. This is promising, because better weather forecasts can reduce renewable energy uncertainty and accelerate decarbonisation. We also compared cyclone tracks from GenCast and ENS with ~100 cyclones observed in 2019. GenCast's ensemble mean cyclone track has a 12-hour position error advantage over ENS out to 4 days, and more actionable track probability fields out to 7 days. Cyclone maximum wind speeds are still generally underestimated (a common problem for ML weather models), but this performance on tracks is really promising. One recent devastating cyclone was Hurricane Milton, which caused >$85 billion in damages. GenCast predicted ~70% probability of landfall in Florida 8.5 days before the hurricane struck (and ~2 days before it even formed). A GenCast ensemble member takes 8 minutes on a TPU chip, versus hours on a supercomputer for physics-based models. This opens up the possibility of large ensembles (eg 1000s of members) which could better estimate risks of extreme events. We don't yet know how much value this will yield over conventional ensemble sizes (~50 members). Like its predecessor (GraphCast), the weights & code of GenCast have been made publicly available: https://lnkd.in/eg78dd7T. We’re looking forward to seeing how the community builds on this! It's been an honour to work on this study led by Ilan Price with such a talented team ✨: Alvaro Sanchez Gonzalez, Ferran Alet Puig, Andrew El-Kadi, Dominic Masters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, Rémi Lam, & Matthew Willson

You might have seen news from our Google DeepMind colleagues lately on GenCast, which is changing the game of weather forecasting by building state-of-the-art weather models using AI. Some of our teams started to wonder – can we apply similar techniques to the notoriously compute-intensive challenge of climate modeling? General circulation models (GCMs) are a critical part of climate modeling, focused on the physical aspects of the climate system, such as temperature, pressure, wind, and ocean currents. Traditional GCMs, while powerful, can struggle with precipitation – and our teams wanted to see if AI could help. Our team released a paper and data on our AI-based GCM, building on our Nature paper from last year - specifically, now predicting precipitation with greater accuracy than prior state of the art. The new paper on NeuralGCM introduces 𝗺𝗼𝗱𝗲𝗹𝘀 𝘁𝗵𝗮𝘁 𝗹𝗲𝗮𝗿𝗻 𝗳𝗿𝗼𝗺 𝘀𝗮𝘁𝗲𝗹𝗹𝗶𝘁𝗲 𝗱𝗮𝘁𝗮 𝘁𝗼 𝗽𝗿𝗼𝗱𝘂𝗰𝗲 𝗺𝗼𝗿𝗲 𝗿𝗲𝗮𝗹𝗶𝘀𝘁𝗶𝗰 𝗿𝗮𝗶𝗻 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀. Kudos to Janni Yuval, Ian Langmore, Dmitrii Kochkov, and Stephan Hoyer! Here's why this is a big deal: 𝗟𝗲𝘀𝘀 𝗕𝗶𝗮𝘀, 𝗠𝗼𝗿𝗲 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆: These new models have less bias, meaning they align more closely with actual observations – and we see this both for forecasts up to 15 days, and also for 20-year projections (in which sea surface temperatures and sea ice were fixed at historical values, since we don’t yet have an ocean model). NeuralGCM forecasts are especially performant around extremes, which are especially important in understanding climate anomalies, and can predict rain patterns throughout the day with better precision. 𝗖𝗼𝗺𝗯𝗶𝗻𝗶𝗻𝗴 𝗔𝗜, 𝗦𝗮𝘁𝗲𝗹𝗹𝗶𝘁𝗲 𝗜𝗺𝗮𝗴𝗲𝗿𝘆, 𝗮𝗻𝗱 𝗣𝗵𝘆𝘀𝗶𝗰𝘀: The model combines a learned physics model with a dynamic differentiable core to leverage both physics and AI methods, with the model trained directly on satellite-based precipitation observations. 𝗢𝗽𝗲𝗻 𝗔𝗰𝗰𝗲𝘀𝘀 𝗳𝗼𝗿 𝗘𝘃𝗲𝗿𝘆𝗼𝗻𝗲! This is perhaps the most exciting news! The team has made their pre-trained NeuralGCM model checkpoints (including their awesome new precipitation models) available under a CC BY-SA 4.0 license. Anyone can use and build upon this cutting-edge technology! https://lnkd.in/gfmAx_Ju 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: Accurate predictions of precipitation are crucial for everything from water resource management and flood mitigation to understanding the impacts of climate change on agriculture and ecosystems. Check out the paper to learn more:  https://lnkd.in/geqaNTRP

The rollout of various new AI weather models over the last year has been something of a blur and, now that the excitement of a cold winter is behind us, we thought it would be time to offer some thoughts from our unique perspective as a leading voice in the energy markets. 1. The AI models are quite useful, but are still not as good, in aggregate, as the better legacy NWP models, especially when looking at fields like 500 mb GPH. Discussions with our operational forecasters, who are in the trenches every day, suggest that the AI models are still used secondarily to the legacy models - "I don't use it other than a gut check/reference". My personal experience is that I still do not consult the AI models nearly as much as a good high-resolution NWP model/ensembles. Perhaps that will evolve with time, but that is the current perspective from those with an extreme level of skin in the game, those who are highly motivated to produce an accurate forecast. 2. However, there are many situations where the legacy models are still severely flawed, especially for 2-meter temperatures, where the AI models add considerable value. We know that the calculation of 2-meter temperatures in the legacy NWP models is a complex process involving highly imperfect parameterizations of surface energy exchanges/fluxes, which is especially complicated and difficult at night. Given that AI models are effectively very mathematically sophisticated analog models, trained on actual observations, they are not crippled by the same biases/errors that the legacy NWP models are. Further, there are certain well-known situations where even the best legacy models do poorly, such as southward-moving shallow and dense cold air masses in the lee of the Rockies and Appalachians, and we've seen multiple instances this past winter where AI models do astoundingly well, while legacy models can be 20-30 degrees off with mistimed cold fronts, etc. 3. The value of AI models relative to legacy models decreases with forecast horizon. An examination of forecast accuracy suggests that AI models can outperform legacy models in the 1-7 day window, but fall off considerably behind that. This applies when comparing both deterministic and ensemble mean solutions. In summary, we are excited to see the continued investment in this space, and are continuing to follow developments as we work to optimally integrate the new models into our product suite. However, we do caution that these new models are a complement, not a replacement, for legacy NWP models, at least for now. #atmosphericg2 #ai #weather

India’s Weather Forecasting Enters a New Era! One month ago, we launched the Bharat Forecast System (BFS), India’s most advanced, fully indigenous weather prediction model. Built by the Indian Institute of Tropical Meteorology (IITM), Pune, it marks a watershed moment in India’s ability to predict and respond to extreme weather events. Over the past 10 years, under the visionary leadership of Hon’ble Prime Minister Sh Narendra Modi , India has made enormous strides in weather and climate science. What was once reliant on foreign models is now led by Indian innovation, more precise, more timely, and more accessible. The Bharat Forecast System is a reflection of this transformation. Here’s what makes it a game-changer: • Double the precision: Earlier models operated at a 12-kilometre resolution. The BFS now works at 6-kilometre resolution, enabling us to predict localised weather conditions with much greater accuracy, especially vital for hilly terrains, coastal zones, and urban areas. • Faster forecasts: BFS runs on India’s new weather supercomputer ARKA, located at IITM Pune. With 11.77 petaflops of computing power and 33 petabytes of storage, ARKA can process billions of weather data points in real time, drastically reducing the time it takes to generate a forecast. • Smarter data integration: BFS uses a blend of data from ISRO satellites, ground stations, ocean buoys, and even information from global partners. This helps the system “see” and simulate weather patterns with far greater clarity. • Open for global science: Unlike many global systems, India is making the BFS data open-access, inviting researchers from around the world to build on our model. It promotes not just national preparedness but global scientific collaboration. What does this mean for the common citizen? Better early warnings before a cyclone hits. More accurate rainfall predictions for farmers. Improved alerts for cloudbursts and flash floods in vulnerable areas. And faster, more targeted disaster response, saving time, resources, and lives. This is not just about supercomputers or models. It is about science serving society, protecting communities, and helping us plan better, from sowing crops to managing cities. With the Bharat Forecast System, India joins the front ranks of nations shaping the future of weather prediction. It is a proud milestone for our scientists and our country, and a strong step forward in building a Viksit Bharat.

Google DeepMind just unveiled GenCast, an AI forecasting model capable of predicting weather patterns up to 15 days in advance. With severe weather events on the rise, this tool could be life-saving—providing critical lead time for disaster preparedness, agriculture, and energy planning. One researcher described GenCast’s impact as “decades’ worth of improvements in a single year.” Its outpu is impressive: it outperformed a leading forecasting model 97.2% of the time. Why it matters for #sustainability professionals: • Better #weather predictions can inform #climaterisk assessments and #resilience strategies. Historical weather models just don't cut it anymore. • AI models like GenCast demonstrate how data and machine learning can help tackle weather-related challenges, paving the way for #innovation in areas like agricultural optimization and supply chain logistics. • These breakthroughs remind us that the intersection of #AI and sustainability isn’t just about tech—it’s about driving impactful solutions for people and the planet. To accelerate collaboration, DeepMind has made GenCast open source, sharing its code and weights to empower researchers worldwide: ➡️ Announcement - "GenCast predicts weather and the risks of extreme conditions with state-of-the-art accuracy" https://lnkd.in/eEVB7v6q ➡️ Academic paper in Nature Magazine - "Probabilistic weather forecasting with machine learning" https://lnkd.in/e7zsmXbZ ➡️ GenCast model code on GitHub - https://lnkd.in/en7DikWk and weights on Google Cloud - https://lnkd.in/eunX7BHC

Nature has just published Microsoft Research's Aurora, the first foundation #model of the #earth system. Aurora outperforms operational #forecasts in predicting #air quality, #ocean waves, tropical #cyclone tracks and high-resolution weather, all at orders of magnitude lower computational cost. Aurora first learns how to generate forecasts through training on #weather patterns from over one #million hours of data. These data are derived from satellites, radar and weather stations, simulations, and forecasts. The model can then be fine-tuned to perform a variety of specific tasks such as predicting wave height or air quality. When #Typhoon Doksuri hit the Philippines in July 2023, the damage was devastating. As reported in Nature, Aurora accurately predicts Typhoon Doksuri’s landfall in the Philippines using measurements from four days in advance of the event (image below). Official predictions at that time mistakenly placed the storm off the coast of Northern Taiwan. Results like this show how #AI is paving the way toward democratizing high-quality climate and weather prediction.  Learn more here: https://lnkd.in/gNiM5tsQ Try it here: https://lnkd.in/gn9DZsry

AI is rapidly improving weather forecasting. Today, Google DeepMind published a new paper in Nature presenting GenCast, a new high resolution ensemble model that includes 50 or more predictions for weather forecasts up to 15 days in advance. GenCast is a diffusion model, similar to generative AI models for image, video, and music generation. However, instead of generating frames or pixels, it produces a probabilistic forecast of atmospheric and weather variables encompassing a range of possible future outcomes. This is more useful than a single prediction for things like extreme weather prediction. And it works well! GenCast is more accurate than the leading operational ensemble model, ECMWF ENS, on 99.8% of forecast variables at lead times greater than 36 hours. https://lnkd.in/gT9-BcNU #AI #weather #google #deepmind

Thrilled to share how my team at Google DeepMind and Google Research is supporting better tropical cyclone prediction with AI 🌀 As we've all seen over the past few years, tropical cyclones cause immense devastation, and improving prediction accuracy is crucial for protecting lives and mitigating economic losses. For me, having family from Florida who've navigated many hurricane seasons, the importance of accurate tropical cyclone prediction is personal. Our new experimental cyclone model offers insights into a cyclone's formation, track, intensity, and more, up to 15 days in advance. Internal testing shows that our model's predictions for cyclone track and intensity are as accurate as, and often more accurate than, current physics-based methods. Curious? You can explore our model's predictions on Weather Lab, our new interactive website for sharing our AI weather models: https://lnkd.in/d6mr2e3i We're collaborating closely with the National Hurricane Center and other leading meteorological organizations to validate and integrate these insights, aiming to support the NHC in providing earlier and more accurate warnings. We are eager for our partners to assess our experimental model's ability to enhance official forecasts this hurricane season. As Wallace Hogsett said to the NYTimes - the union of skilled human forecasters and A.I. tools has the potential to create “a really powerful partnership.” This is an exciting step towards more accurate and earlier cyclone predictions, which can truly save lives and protect communities. Learn more in our blog post: https://lnkd.in/dvcYZD5S Or check out this story in the New York Times: https://lnkd.in/dUvtadBH

AI is cutting weather forecast times from hours to under a minute, while matching or beating traditional models on accuracy. That kind of leap matters. Google’s GraphCast and Huawei’s Pangu-Weather are only a few examples of how machine learning can process massive datasets faster and more efficiently than physics-based systems. But why does it matter? During a major hurricane or flood, even a few hours of early warning can save lives and protect infrastructure. Faster forecasts mean earlier evacuations, better emergency planning, and smarter decisions on the ground. For industries that depend on weather—agriculture, energy, logistics, emergency response—this shift isn’t just interesting, it’s operationally critical. As climate volatility grows, the ability to forecast extreme events faster and more precisely could make all the difference. You can read more about it here: https://lnkd.in/eQi-YY-F #AI #WeatherForecasting #ClimateTech