Advanced Tools for Climate Scenario Modeling

Further progress in AI+climate modeling "Applying the ACE2 Emulator to SST Green's Functions for the E3SMv3 Global Atmosphere Model". Building on ACE2 model which uses our spherical Fourier neural operator (SFNO) architecture, this work shows that ACE2 can replicate climate model responses to sea surface temperature perturbations with high fidelity at a fraction of the cost. This accelerates climate sensitivity research and helps us better understand radiative feedbacks in the Earth system. Background: The SFNO architecture was first used in training FourCastNet weather model, whose latest version (v3) has state-of-art probabilistic calibration. AI+Science is not just about blindly applying the standard transformer/CNN "hammer". It is about carefully designing neural architectures that incorporate domain constraints like geometry and multiple scales, while being expressive and easy to train. SFNO accomplishes both: it incorporates multiple scales, and it respects the spherical geometry and this is critical for success in climate modeling. Unlike short-term weather, which requires only a few autoregressive steps for rollout, climate modeling requires long rollouts with thousands or even greater number of time steps. All other AI-based models fail for long-term climate modeling including Pangu and GraphCast which ignore the spherical geometry. Distortions start building up at the poles since the models assume domain is a rectangle, and they lead to catastrophic failures. Structure matters in AI+Science!

🌍Typology of climate scenarios by MSCI Inc. and United Nations Environment Programme Finance Initiative (UNEP FI) 🔹 1. Fully Narrative Scenarios These scenarios are qualitative descriptions of potential climate futures. ✅ Strengths: - Easily customizable - Useful for high-level strategic discussions - Can capture complex risks that are difficult to quantify ⚠️ Limitations: - Subjective and vulnerable to bias - Lack of numerical outputs makes them hard to integrate into risk models 🔍 Example Providers: - University of Exeter & Universities Superannuation Scheme 🔹 2. Quantified Narrative Scenarios This type builds on fully narrative scenarios by adding expert-driven quantitative estimates (macroeconomic forecasts, asset class returns, regional physical risks). ✅ Strengths: - Balances qualitative storytelling with numerical data - Allows for scenario comparisons without requiring sophisticated models - Easier to communicate results with clear quantitative insights ⚠️ Limitations: - Can give a false sense of precision if assumptions are weak - Still dependent on subjective expert input, leading to potential biases 🔍 Example Providers: - MSCI Sustainability Institute & University of Exeter – Estimating physical climate risks based on expert-defined damage functions. - IEA - WEO 🔹 3. Model-Driven Scenarios These scenarios rely on integrated quantitative models to project how climate change and transition risks might evolve under different policy and economic conditions, using macroeconomic models, IAMs, and energy system models. ✅ Strengths: Highly structured and data-driven, reducing subjectivity. Can produce detailed, sector-specific outputs useful for investment decisions. Widely used by regulators and financial institutions for stress testing. ⚠️ Limitations: - Expensive and time-consuming to develop and maintain - “Black box” nature of complex models makes interpretation difficult - Results are only as good as underlying assumptions and data inputs 🔍 Example Providers: - NGFS – Climate scenarios for central banks and financial supervisors - IEA – Net-Zero Emissions by 2050, STEPS & APS scenarios - IPCC – SSPs & RCPs 🔹 4. Probabilistic Scenarios Probabilistic models go beyond single-scenario forecasting by assigning probabilities, variance, and uncertainty estimates to different climate outcomes. ✅ Strengths: - Models uncertainty, improving risk management - Enables sophisticated stress testing for asset prices, portfolios, and corporate exposure - Valuable for insurance, catastrophe modeling, and financial risk assessments ⚠️ Limitations: - Highly complex and computationally demanding - Requires strong assumptions about uncertainty - Limited research on how climate change affects probability distributions 🔍 Example Providers: - NGFS & IPCC Probabilistic Models #ClimateFinance #ScenarioAnalysis #SustainableInvesting #RiskManagement #climatescenarios

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

In this week's column, I look at NVIDIA's new generative foundation model that it says enables simulations of Earth’s global climate with an unprecedented level of resolution. As is so often the case with powerful new technology, however, the question is what else humans will do with it. The company expects that climate researchers will build on top of its new AI-powered model to make climate predictions that focus on five-kilometer areas. Previous leading-edge global climate models typically don’t drill below 25 to 100 kilometers. Researchers using the new model may be able to predict conditions decades into the future with a new level of precision, providing information that could help efforts to mitigate climate change or its effects. A 5-kilometer resolution may help capture vertical movements of air in the lower atmosphere that can lead to certain kinds of thunderstorms, for example, and that might be missed with other models. And to the extent that high-resolution near-term forecasts are more accurate, the accuracy of longer-term climate forecasts will improve in turn, because the accuracy of such predictions compounds over time. The model, branded by Nvidia as cBottle for “Climate in a Bottle,” compresses the scale of Earth observation data 3,000 times and transforms it into ultra-high-resolution, queryable and interactive climate simulations, according to Dion Harris, senior director of high-performance computing and AI factory solutions at Nvidia. It was trained on high-resolution physical climate simulations and estimates of observed atmospheric states over the past 50 years. It will take years, of course, to know just how accurate the model’s long-term predictions turn out to be. The The Alan Turing Institute of AI and the Max Planck Institute of Meteorology, are actively exploring the new model, Nvidia said Tuesday at the ISC 2025 computing conference in Hamburg. Bjorn Stevens, director of the Planck Institute, said it “represents a transformative leap in our ability to understand, predict and adapt to the world around us.” The Earth-2 platform is in various states of deployment at weather agencies from NOAA: National Oceanic & Atmospheric Administration in the U.S. to G42, an Abu Dhabi-based holding company focused on AI, and the National Science and Technology Center for Disaster Reduction in Taiwan. Spire Global, a provider of data analytics in areas such as climate and global security, has used Earth-2 to help improve its weather forecasts by three orders of magnitude with regards to speed and cost over the last three or four years, according to Peter Platzer, co-founder and executive chairman.

The Nature Communications article "How to stop being surprised by unprecedented weather" outlines a comprehensive framework to anticipate and manage the risks of extreme, previously unobserved weather events. The article’s central thesis is that surprise should not be the default response to such events—and that science, policy, and disaster planning can work in concert to build resilience. These methods help anticipate extreme weather events beyond what has occurred in the observational record: a. Conventional Statistical Methods - Use historical weather data and extreme value theory to estimate probabilities of rare events. Limitations: Short observational records, underestimation of extremes, and inability to simulate events beyond past climate conditions. b. Past Events and Proxy Data - Extend the view of climate risk through historical documentation, oral history, and paleoclimate proxies (tree rings, sediments, etc.). Benefits: Reveal long-term variability and past extremes that modern records miss. Limitations: Coarse resolution, dating uncertainty, and difficulty aligning with present-day conditions. c. Event-Based Storylines - Construct physically plausible scenarios of specific high-impact events using counterfactuals and modeling. Useful for local decision-making and public engagement. Limitations: Focused on specific events, often non-probabilistic, and dependent on expert input. d. Weather and Climate Model Data Exploration Mine large ensembles of model outputs (e.g., UNSEEN, SMILEs, CORDEX) for unobserved but plausible extremes. Enables exploration of events outside the observational record using physical consistency. Limitations: Computationally intensive, resolution trade-offs, and model biases.

𝗔𝗜-𝗣𝗼𝘄𝗲𝗿𝗲𝗱 𝗕𝗿𝗲𝗮𝗸𝘁𝗵𝗿𝗼𝘂𝗴𝗵 𝗶𝗻 𝗖𝗹𝗶𝗺𝗮𝘁𝗲 𝗠𝗼𝗱𝗲𝗹𝗶𝗻𝗴 A new machine learning based approach called NeuralGCM has been developed by Google AI researchers in partnership with European Centre for Medium-Range Weather Forecasts. This model combines traditional physics-based modeling with machine learning to create more accurate and efficient climate simulations. NeuralGCM can simulate weather events from days to weeks, and climate trends over decades. This is a significant breakthrough in climate modeling, and it has the potential to improve our understanding of climate change. With more accurate simulations, we can better predict the impacts of climate change and develop more effective mitigation and adaptation strategies. . Here are some of the key benefits of NeuralGCM: 🎯 More accurate simulations: NeuralGCM can simulate weather events and climate trends with greater accuracy than traditional models. ⏱️ More efficient simulations: NeuralGCM is much faster and cheaper to run than traditional models. 🌍 More comprehensive simulations: NeuralGCM can be expanded to include other parts of the climate system, such as the oceans and the biosphere. https://lnkd.in/g722iYSp

We know the Earth is getting warmer, but not what it means specifically for different regions. To figure this out, scientists do climate modelling. 🔎 🌍 , Google Research has published groundbreaking advancements in climate prediction using the power of #AI! Typically, researchers use "climate modelling" to understand the regional impacts of climate change, but current approaches have large uncertainty. Introducing NeuralGCM: a new atmospheric model that outperforms existing models by combining AI with physics-based modelling for improved accuracy and efficiency. Here’s why it stands out: ✅ More Accurate Simulations When predicting global temperatures and humidity for 2020, NeuralGCM had 15-50% less error than the state-of-the-art model "X-SHiELD". ✅ Faster Results NeuralGCM is 3,500 times quicker than X-SHiELD. If researchers simulated a year of the Earth's atmosphere with X-SHiELD, it would take 20 days to complete —  whereas NeuralGCM achieves this in just 8 minutes. ✅ Greater Accessibility Google Research has made NeuralGCM openly available on GitHub for non-commercial use, allowing researchers to explore, test ideas, and improve the model’s functionality. The research showcases AI’s ability to help deliver more accurate, efficient, and accessible climate predictions, which is critical to navigating a changing global climate. Read more about the team’s groundbreaking research in   Nature Portfolio’s  latest article! → https://lnkd.in/e-Etb_x4 #AIforClimateAction #Sustainability #AI

AI has the potential to bring new waves of innovation, social and economic progress on a scale we’ve not seen before - including supercharging scientific progress. This week, Google published NeuralGCM: an openly available tool for fast, accurate climate modelling - critical to a changing global climate. We know that the Earth is getting warmer, but it’s hard to predict what that means for each different region. To figure this out, scientists use climate modelling. But current approaches have large uncertainty, including systematic errors - like forecasting extreme rain that is only half as intense as what scientists actually observe. That’s where NeuralGCM comes in. It combines physics-based modelling and AI to simulate the Earth’s atmosphere - making it faster and more accurate than existing climate models. For scientists exploring how to build better weather and climate models, it should make a huge difference in helping them understand the effects of the climate crisis on our world - and it could also be great for meteorologists making predictions about our daily weather! Interested in learning more? Read all about it here and watch the video below ⬇️ https://lnkd.in/e_bCuAhq 

𝗔𝗜 𝗳𝗼𝗿 𝗚𝗢𝗢𝗗: 𝗡𝗔𝗦𝗔 𝗮𝗻𝗱 𝗜𝗕𝗠 𝗹𝗮𝘂𝗻𝗰𝗵 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲 𝗔𝗜 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹 𝗳𝗼𝗿 𝗺𝗼𝗿𝗲 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝘄𝗲𝗮𝘁𝗵𝗲𝗿 𝗮𝗻𝗱 𝗰𝗹𝗶𝗺𝗮𝘁𝗲 𝗳𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴! 🌍 (𝗧𝗵𝗶𝘀 𝗶𝘀 𝘄𝗵𝗮𝘁 𝘀𝗵𝗼𝘂𝗹𝗱 𝗴𝗲𝘁 𝗺𝗼𝗿𝗲 𝘀𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗽𝗹𝗲𝗮𝘀𝗲 𝗮𝗻𝗱 𝗡𝗢𝗧 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗪𝗿𝗮𝗽𝗽𝗲𝗿!) In collaboration with NASA, IBM just launched Prithvi WxC an open-source, general-purpose AI model for weather and climate-related applications. And the truly remarkable part is that this model can run on a desktop computer. 𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘆𝗼𝘂 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗸𝗻𝗼𝘄: ⬇️ → The Prithvi WxC model (2.3-billion parameter) can create six-hour-ahead forecasts as a “zero-shot” skill – meaning it requires no tuning and runs on readily available data. → This AI model is designed to be customized for a variety of weather applications, from predicting local rainfall to tracking hurricanes or improving global climate simulations. → The model was trained using 40 years of NASA’s MERRA-2 data and can now be quickly tuned for specific use cases. And unlike traditional climate models that require massive supercomputers, this one operates on a desktop. Uniqueness lies in the ability to generalize from a small, high-quality sample of weather data to entire global forecasts. → This AI-powered model outperforms traditional numerical weather prediction methods in both accuracy and speed, producing global forecasts up to 10 days in advance within minutes instead of hours. → This model has immense potential for various applications, from downscaling high-resolution climate data to improving hurricane forecasts and capturing gravity waves. It could also help estimate the extent of past floods, forecast hurricanes, and infer the intensity of past wildfires from burn scars. It will be exciting to see what downstream apps, use cases, and potential applications emerge. What’s clear is that this AI foundation model joins a growing family of open-source tools designed to make NASA’s vast collection of satellite, geospatial, and Earth observational data faster and easier to analyze. With decades of observations, NASA holds a wealth of data, but its accessibility has been limited — until recently. This model is a big step toward democratizing data and making it more accessible to all. 𝗔𝗻𝗱 𝘁𝗵𝘀 𝗶𝘀 𝘆𝗲𝘁 𝗮𝗻𝗼𝘁𝗵𝗲𝗿 𝗽𝗿𝗼𝗼𝗳 𝘁𝗵𝗮𝘁 𝘁𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗔𝗜 𝗶𝘀 𝗼𝗽𝗲𝗻, 𝗱𝗲𝗰𝗲𝗻𝘁𝗿𝗮𝗹𝗶𝘇𝗲𝗱, 𝗮𝗻𝗱 𝗿𝘂𝗻𝗻𝗶𝗻𝗴 𝗮𝘁 𝘁𝗵𝗲 𝗲𝗱𝗴𝗲. 🌍 🔗 Resources: Download the models from the Hugging Face repository: https://lnkd.in/gp2zmkSq Blog post: https://ibm.co/3TDul9a Research paper: https://ibm.co/3TAILXG #AI #ClimateScience #WeatherForecasting #OpenSource #NASA #IBMResearch

💡 A Practical Guide to Climate Scenarios! Really pleased to have written the forward to this valuable report on the types and applications of climate scenarios by MSCI Inc. and my former United Nations Environment Programme Finance Initiative (UNEP FI) FI colleagues Looking for a handy summary of the types of scenarios from qualitative to quantitative? Here it is: 1. 𝗙𝘂𝗹𝗹𝘆 𝗡𝗮𝗿𝗿𝗮𝘁𝗶𝘃𝗲 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 These scenarios are qualitative descriptions of potential climate futures. ✅ Strengths: - Easily customizable - Useful for high-level strategic discussions - Can capture complex risks that are difficult to quantify ⚠️ Limitations: - Subjective and vulnerable to bias - Lack of numerical outputs makes them hard to integrate into risk models 2. 𝗤𝘂𝗮𝗻𝘁𝗶𝗳𝗶𝗲𝗱 𝗡𝗮𝗿𝗿𝗮𝘁𝗶𝘃𝗲 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 This type builds on fully narrative scenarios by adding expert-driven quantitative estimates (macroeconomic forecasts, asset class returns, regional physical risks). ✅ Strengths: - Balances qualitative storytelling with numerical data - Allows for scenario comparisons without requiring sophisticated models - Easier to communicate results with clear quantitative insights ⚠️ Limitations: - Can give a false sense of precision if assumptions are weak - Still dependent on subjective expert input, leading to potential biases 3. 𝗠𝗼𝗱𝗲𝗹-𝗗𝗿𝗶𝘃𝗲𝗻 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 These scenarios rely on integrated quantitative models to project how climate change and transition risks might evolve under different policy and economic conditions, using macroeconomic models, IAMs, and energy system models. ✅ Strengths: Highly structured and data-driven, reducing subjectivity. Can produce detailed, sector-specific outputs useful for investment decisions. Widely used by regulators and financial institutions for stress testing. ⚠️ Limitations: - Expensive and time-consuming to develop and maintain - “Black box” nature of complex models makes interpretation difficult - Results are only as good as underlying assumptions and data inputs 4. 𝗣𝗿𝗼𝗯𝗮𝗯𝗶𝗹𝗶𝘀𝘁𝗶𝗰 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 Probabilistic models go beyond single-scenario forecasting by assigning probabilities, variance, and uncertainty estimates to different climate outcomes. ✅ Strengths: - Models uncertainty, improving risk management - Enables sophisticated stress testing for asset prices, portfolios, and corporate exposure - Valuable for insurance, catastrophe modeling, and financial risk assessments ⚠️ Limitations: - Highly complex and computationally demanding - Requires strong assumptions about uncertainty - Limited research on how climate change affects probability distributions #ClimateFinance #ClimateScenarios #SustainableInvesting #RiskManagement #ScenarioAnalysis #Risk #Finance