Why current climate models lack local accuracy

The challenges of climate change modeling: "The Earth is an unfathomably complex place, a nesting doll of systems within systems. Feedback loops among temperature, land, air, and water are made even more complicated by the fact that every place on Earth is a little different. Natural variability and human-driven warming further alter the rules that govern each of those fundamental interactions. On every continent except Antarctica, certain regions showed up as mysterious hot spots, suffering repeated heat waves worse than what any model could predict or explain. Across places where a third of humanity lives, actual daily temperature records are outpacing model predictions. And a global jump in temperature that lasted from mid-2023 to this past June remains largely unexplained. Per one researcher: “We have to approximate cloud formation because we don’t have the small scales necessary to resolve individual water droplets coming together." "Similarly, models approximate topography, because the scale at which mountain ranges undulate is smaller than the resolution of global climate models, which tend to represent Earth in, at best, 100-square-kilometer pixels. That resolution is good for understanding phenomena such as Arctic warming over decades. But “you can’t resolve a tornado worth anything.” "Models simply can’t function on the scale at which people live, because assessing the impact of current emissions on the future world requires hundreds of years of simulations. Some variables are missing from climate models entirely. Trees and land have been considered major sinks for carbon emissions. But it is changing: Trees and land absorbed much less carbon than normal in 2023. In Finland, forests have stopped absorbing the majority of the carbon they once did, and recently became a net source of emissions, which swamped all gains the country has made in cutting emissions from all other sectors since the early 1990s. The interactions of the ice sheets with the oceans are also largely missing from models. Changing ocean-temperature patterns are currently making climate modelers at NOAA rethink their models of El Niño and La Niña; the agency initially predicted that La Niña’s cooling powers would kick in much sooner than it now appears they will. "The models may be underestimating future climate risks across several regions because of a yet-unclear limitation. And underestimating risk is far more dangerous than overestimating it. Excerpts from The Atlantic article: Climate Models Can’t Explain What’s Happening to Earth Global warming is moving faster than the best models can keep a handle on. By Zoë Schlanger

“Fifty years into the project of modeling Earth’s future climate, we still don’t really know what’s coming. Some places are warming with more ferocity than expected. Extreme events are taking scientists by surprise. Right now, as the bald reality of climate change bears down on human life, scientists are seeing more clearly the limits of our ability to predict the exact future we face. The coming decades may be far worse, and far weirder, than the best models anticipated… This is a problem. The world has warmed enough that city planners, public-health officials, insurance companies, farmers, and everyone else in the global economy want to know what’s coming next for their patch of the planet… Today’s climate models very accurately describe the broad strokes of Earth’s future. But warming has also now progressed enough that scientists are noticing unsettling mismatches between some of their predictions and real outcomes… Across places where a third of humanity lives, actual daily temperature records are outpacing model predictions… And a global jump in temperature that lasted from mid-2023 to this past June remains largely unexplained… Trees and land are major sinks for carbon emissions, and that this fact might change is not accounted for in climate models. But it is changing: Trees and land absorbed much less carbon than normal in 2023, according to research published last October… The interactions of the ice sheets with the oceans are also largely missing from models, Schmidt told me, despite the fact that melting ice could change ocean temperatures, which could have significant knock-on effects… The models may be underestimating future climate risks across several regions because of a yet-unclear limitation. And, Rohde said, underestimating risk is far more dangerous than overestimating it.” #ClimateRisk #TransitionRisk https://lnkd.in/eiSRvUeF

🌍 Reality vs. Climate Models 🌍 Climate models are falling behind reality. New research reveals that real-world data shows more extreme and unexpected climate changes than models predict. Why? Most models can’t simulate the small-scale processes that run Earth’s climate systems, like: 🌪️ Jet streams ☁️ Cloud formation 🌡️ Soil moisture interactions 🌊 Ocean currents These tiny, daily processes combine in complex ways that models can’t fully capture. This means: • Extreme events (like heat, storms, and marine heatwaves) are happening faster and more intensely than models predict. • The interconnected systems of land, atmosphere, and oceans are reacting in non-linear and amplifying ways we didn’t expect. 🚨 A key finding: Climate models underestimate extreme trends by up to 4x in some areas. 👉 What this means. 1️⃣ Our understanding of the climate crisis must evolve—models need to account for these overlooked processes. 2️⃣ Rapid emissions cuts are essential to avoid more surprises from an accelerating climate system. The gap between models and reality is growing. The science is clear: the climate is changing faster than we are prepared for. 📖 Study: “Global emergence of regional heatwave hotspots outpaces climate model simulations” Link to research: https://lnkd.in/dABZVEry

Thrilled to unveil our latest work: multi-modal machine learning to forecast localized weather! We construct a graph neural network to learn dynamics at point locations, where typical gridded forecasts miss significant variation. Paper: https://lnkd.in/eBmfsJin Weather dataset: https://lnkd.in/ejCG8bKs Code: https://lnkd.in/eQg-JzQJ AI weather models have made huge strides, but most still emulate products like ERA5, which struggle to capture near-surface wind dynamics. The correlation between ERA5 and ground weather station data is low due to topography, buildings, vegetation, and other local factors. In this work, we forecast near-surface wind at localized off-grid locations using a message-passing graph neural network ("MPNN"). The graph is heterogeneous, integrating both global forecasts (ERA5) and historical local weather station data as different nodes. What do we find? First off, ERA5 interpolation performs poorly, failing to capture local wind variations, especially in coastal and inland regions with complex conditions. An MLP trained on historical data at a location performs better than ERA5 interpolation, as it learns from the station's past observations. However, it struggles with longer lead times and lacks the spatial context necessary to capture weather patterns. Meanwhile, our MPNN dramatically improves performance, reducing the error by over 50% compared to the MLP. This is because the MPNN incorporates spatial information through message passing, allowing it to learn local weather dynamics from both station data and global forecasts. Interestingly, adding ERA5 data to the MLP does not improve its performance significantly. The MLP struggles to integrate spatial information from global forecasts, while the MPNN excels, highlighting the importance of combining global and local data. Large improvements in forecast accuracy occur at both coastal and inland locations. Our model shows a 92% reduction in MSE relative to ERA5 interpolation overall. This work showcases the strength of machine learning in combining multi-modal data. By using a graph to integrate global and local weather data, we were able to generate much more accurate localized weather forecasts! Congrats to Qidong Yang and Jonathan Giezendanner for the great work, and thanks to Campbell Watson, Daniel Salles Chevitarese, Johannes Jakubik, Eric Schmitt, Anirban C., Jeremy Vila, Detlef Hohl, and Chris Hill for a wonderful collaboration. Thanks also to our partners at Amazon Web Services (AWS) for providing cloud computing and technical support!

Every year, natural disasters hit harder and closer to home. But when city leaders ask, "How will rising heat or wildfire smoke impact my home in 5 years?"—our answers are often vague. Traditional climate models give sweeping predictions, but they fall short at the local level. It's like trying to navigate rush hour using a globe instead of a street map. That’s where generative AI comes in. This year, our team at Google Research built a new genAI method to project climate impacts—taking predictions from the size of a small state to the size of a small city. Our approach provides: - Unprecedented detail – in regional environmental risk assessments at a small fraction of the cost of existing techniques - Higher accuracy – reduced fine-scale errors by over 40% for critical weather variables and reduces error in extreme heat and precipitation projections by over 20% and 10% respectively - Better estimates of complex risks – Demonstrates remarkable skill in capturing complex environmental risks due to regional phenomena, such as wildfire risk from Santa Ana winds, which statistical methods often miss Dynamical-generative downscaling process works in two steps: 1) Physics-based first pass: First, a regional climate model downscales global Earth system data to an intermediate resolution (e.g., 50 km) – much cheaper computationally than going straight to very high resolution. 2) AI adds the fine details: Our AI-based Regional Residual Diffusion-based Downscaling model (“R2D2”) adds realistic, fine-scale details to bring it up to the target high resolution (typically less than 10 km), based on its training on high-resolution weather data. Why does this matter? Governments and utilities need these hyperlocal forecasts to prepare emergency response, invest in infrastructure, and protect vulnerable neighborhoods. And this is just one way AI is turbocharging climate resilience. Our teams at Google are already using AI to forecast floods, detect wildfires in real time, and help the UN respond faster after disasters. The next chapter of climate action means giving every city the tools to see—and shape—their own future. Congratulations Ignacio Lopez Gomez, Tyler Russell MBA, PMP, and teams on this important work! Discover the full details of this breakthrough: https://lnkd.in/g5u_WctW  PNAS Paper: https://lnkd.in/gr7Acz25

While my focus these days is 100% on Earthmover, there are still a couple of papers in the pipeline from my old lab at Lamont-Doherty Earth Observatory. This one just published in GRL, led by the amazing duo Julius Busecke and Dhruv Balwada addresses an important issue in climate modeling: the impact of model resolution on air-sea fluxes: https://lnkd.in/guydiNyZ The exchange of heat between the ocean and atmosphere plays a crucial role in shaping Earth's weather and climate. However, many climate models miss the effects of small-scale flow features like sharp spatial temperature differences in the ocean or local changes in wind. In this study, we used high-resolution coupled climate simulations to show that these small-scale variations have a big impact on how much heat moves from the ocean to the atmosphere. On average, they lead to more ocean cooling—about 4 watts per square meter across the globe—and in some regions, the effect is much stronger, reaching up to 100 watts per square meter. The two main drivers of these changes are small-scale fluctuations in wind and ocean temperatures. Our results suggest that improving how climate models represent these small-scale features could make them more accurate at predicting ocean heat uptake. In addition to the science result, I'm especially proud of the open-science aspect of this paper. Doing this analysis was very data-intensive, requiring us to reprocess dozens of terabytes of model output with expensive bulk-formula and diffusion-smoothing algorithms. The code to reproduce is not just available--it actually works! 😂 And all of the data is publicly available via high-throughput cloud object storage. Check out the github repo for all the details of how we did it: https://lnkd.in/gA99y_gG

Although nearly a month has passed since the heavy precipitation event that struck the #Valencia region, I believe it’s still relevant to analyze some key aspects of this occurrence. This time, rather than examining the historical data from nearby stations (https://lnkd.in/emavXauX), I want to focus on the spatial distribution of accumulated precipitation, particularly in the context of predictability. Obtaining official data from AEMET is a challenge, but I’ve compiled observations from various weather networks for this analysis. In the plot, each point represents a station and is color-coded to indicate the total precipitation recorded on October 29th, with colors ranging from white to yellow, blue, and red. The area with the highest precipitation was centered between #Turis and #Chiva, where values reached up to 640 mm. The official AEMET station in Turis even recorded an astounding 771 mm. In stark contrast, Valencia itself only recorded about 10 mm—an extraordinary difference given the mere 20 km separating these locations. Now you may ask yourself what are those triangles in the background. The smaller triangles represent the grid of ICON-EU, one of Europe’s state-of-the-art weather forecasting models. Since this model assigns only a single value to each grid cell (represented by a triangle), it’s evident that it cannot accurately capture the extremes of events like this one, where there are often multiple stations within a single cell. But that's not the end of the story. The model's “nominal” resolution—its grid spacing—is not a true reflection of the scale of phenomena it can resolve. To better understand this, think of pixels in an image: accurately detecting and representing the shape of an object requires a sufficient number of pixels. Similarly, for weather models, the “effective” resolution is typically 4 to 10 times coarser than the nominal resolution. In the plot, this limitation is illustrated by the larger triangles. As this event demonstrates, such highly localized phenomena are challenging to capture, even with relatively high resolutions. Accurately forecasting events like this requires increasingly high-resolution models with sub-kilometer grid spacing. However, even with such advancements, we remain constrained by uncertainties in the initial conditions, which propagate exponentially over time.

We rely heavily on global models to understand climate change impacts on biodiversity. 𝗕𝘂𝘁 𝗺𝗼𝗱𝗲𝗹𝘀 𝗮𝗿𝗲 𝘀𝗶𝗺𝗽𝗹𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 𝗼𝗳 𝗿𝗲𝗮𝗹𝗶𝘁𝘆, 𝗮𝗻𝗱 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗴𝗿𝗼𝘂𝗻𝗱-𝘁𝗿𝘂𝘁𝗵𝗶𝗻𝗴, 𝘁𝗵𝗲𝘆 𝗰𝗮𝗿𝗿𝘆 𝘀𝗶𝗴𝗻𝗶𝗳𝗶𝗰𝗮𝗻𝘁 𝗿𝗶𝘀𝗸𝘀. While remote sensing and ecological modeling are powerful tools, they often miss the subtle, place-based changes that occur in real time. A recent review by Cruz-Gispert et al. (2025) in Conservation Biology highlights the magnitude of this disconnect. The study analyzed local observations from Indigenous peoples and local communities worldwide. The comparison between what people see on the ground and what global databases record is insightful: 📉 The blind spot The authors found that Indigenous and local communities 𝗿𝗲𝗽𝗼𝗿𝘁𝗲𝗱 𝗰𝗹𝗶𝗺𝗮𝘁𝗲 𝗰𝗵𝗮𝗻𝗴𝗲 𝗶𝗺𝗽𝗮𝗰𝘁𝘀 𝗼𝗻 𝟱𝟭𝟱 𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝘀𝗽𝗲𝗰𝗶𝗲𝘀. 41% of these species have not even been evaluated by the IUCN Red List. More concerningly, of the species that have been evaluated, 85% are not currently listed as threatened by climate change in the global database. While the IUCN assessment might say a species is "Least Concern," 𝗽𝗲𝗼𝗽𝗹𝗲 𝗼𝗻 𝘁𝗵𝗲 𝗴𝗿𝗼𝘂𝗻𝗱 𝗮𝗿𝗲 𝗼𝗯𝘀𝗲𝗿𝘃𝗶𝗻𝗴 𝗰𝗵𝗮𝗻𝗴𝗲𝘀 𝗶𝗻 𝗶𝘁𝘀 𝗽𝗵𝗲𝗻𝗼𝗹𝗼𝗴𝘆, 𝗮𝗯𝘂𝗻𝗱𝗮𝗻𝗰𝗲, 𝗼𝗿 𝗺𝗼𝗿𝗽𝗵𝗼𝗹𝗼𝗴𝘆 (like fat content in marine mammals or the taste of wild fruits). ⚠️ The language gap It is important to note a limitation: this review focused on literature published in English. However, rather than invalidating the results, this suggests the gap is likely wider. If we were to include local knowledge documented in other languages, the divergence between global models and local reality would arguably be even more pronounced. 🌱 The human sensor The most brilliant researcher I have ever had the pleasure of working with (distinguished not just for her scientific rigor but for her human quality) once told me: "I have been researching these alpine grasslands for 20 years, year after year. There are things I see and observe that I simply cannot capture with standard methods or present in a paper." We cannot afford to lose the connection with the local scale. We need to re-value field-based methodologies that listen to the territory. Without them, our global maps will remain incomplete. 👇 References Cruz-Gispert et al. (2025) https://lnkd.in/e3RcQD4h #IndigenousKnowledge #ClimateChange #Biodiversity #EcologicalMonitoring #SciencePolicy #FieldWork #SocialEcologicalSystems

Good morning Meteorologists and Atmospheric Scientists around the globe, Small national #weather services face a unique challenge. They are responsible for protecting lives and livelihoods with forecasts that must be highly local, yet most of the global weather models they rely on were never designed to resolve weather at the scale of small #countries. Take #Eswatini as an example. Like many smaller nations, much of its terrain, river basins, and population centers can fall within just a handful of grid points in a global model. With typical global grid spacing around 25 km by 25 km, critical local features such as mountains, valleys, and land–lake or land–sea contrasts are often smoothed out. This makes it difficult to accurately forecast localized heavy #rainfall, terrain-driven wind patterns, flash #flooding, or convective #storms that can vary dramatically over short distances. This is where the NSF NCAR - The National Center for Atmospheric Research's Weather Research and Forecasting (#WRF) model becomes a powerful tool for national weather services. WRF allows meteorologists to take the big-picture guidance from global models and dynamically downscale it to much higher resolution over their own country or region. A single 25 km global grid cell can be refined into grids of 5 km, 3 km, or even finer, revealing details that matter on the ground. With this increased resolution, forecasters can better capture orographic rainfall, localized #thunderstorms, valley temperature gradients, and wind acceleration through mountain passes, all of which are especially important in smaller, topographically diverse nations. Beyond the model itself, the WRF ecosystem is a major advantage. Using WRF-Python, forecasters and researchers can efficiently analyze and visualize a wide range of variables directly from model output. Rainfall accumulation, instability indices, wind shear, vertical motion, #temperature anomalies, and many other fields can be plotted, compared, and communicated clearly to decision-makers. This improves not only forecast accuracy, but also forecast confidence and communication. For national meteorological services in smaller countries, running WRF locally is not about replacing global models. It is about enhancing them. By combining global guidance with high-resolution local modeling, services like the Eswatini National Weather Service, and many others across #Africa, the #Caribbean, the #Pacific, and beyond, can deliver forecasts that are more relevant, more actionable, and more impactful for their communities. Investing in local high-resolution modeling is an investment in resilience, preparedness, and better climate and weather services for the people who depend on them every day.

Bridging the gap between global weather models and local reality Accurate weather forecasts at specific locations are critical for wildfire management, renewable energy, agriculture, and infrastructure planning. Yet most numerical and AI-based weather models still operate on coarse grids, systematically missing near-surface, local effects — especially for wind. A recent study, “Local Off-Grid Weather Forecasting With Multi-Modal Earth Observation Data”, demonstrates a powerful alternative: instead of simply downscaling gridded forecasts, the authors correct large-scale weather predictions using direct station-level observations. By combining: • historical measurements from weather stations • large-scale numerical forecasts (ERA5 / HRRR) • and a transformer model with dynamic spatial attention the approach delivers highly accurate off-grid forecasts at irregular station locations. The results are striking — up to 80% error reduction for near-surface wind compared to gridded forecasts alone. The key takeaway: Even the best global or ML weather models cannot achieve local accuracy without direct station inputs. Transformers excel here because they can dynamically learn which nearby observations matter most under changing conditions. This work points toward a future where local weather intelligence is no longer limited by grid resolution — enabling better decisions in high-stakes, location-sensitive applications. Source: https://lnkd.in/dp5bSKWa #WeatherForecasting #EarthObservation #AI #Transformers #ClimateTech #RenewableEnergy #WildfireManagement #MachineLearning