Climate scientists are confronting a hard truth: some of the most widely used models are struggling to keep up with the pace and texture of real‑world warming. The physics at their core remains sound, but gaps in data and missing processes are leaving them blind to crucial details, especially around extremes. The story is not that the science failed, but that the planet is now testing models in ways their creators never fully anticipated.
As warming accelerates, the mismatch between projections and observations is sharpening into a practical problem for governments, insurers, and communities trying to plan for the next decade. The missing pieces range from how forests absorb carbon to how the Arctic and tropical Pacific feed back into mid‑latitude storms. When I look across the latest research, the pattern is clear: climate models got key aspects wrong because the data they needed either did not exist, or sat outside the resolution and scope of the tools we built.
Big picture: models that work, until they don’t
At the broadest scale, today’s climate models still capture the main arc of global warming, including the rise in average temperatures and the link to greenhouse gas emissions. Researchers like Jan have emphasized that today these tools very accurately describe the broad strokes of Earth’s future, which is why long‑term projections of several degrees of warming remain robust. The problem emerges when I zoom in, either in space or time, and ask the models to explain why a particular month, region, or type of extreme is behaving so wildly.
That tension is now unavoidable because warming has progressed far enough that the climate system is throwing up events outside the historical range that models were tuned against. Jan notes that as simulations run up against the reality of dramatic climate change, some of their limitations are being exposed in sectors that have been relatively understudied since the early 1990s, a point underscored in work that begins, fittingly, with the phrase At the. I see a widening gap between what the models were built to do, and what policymakers now need from them: not just a sense of the century, but a sharper view of the next season, the next storm, the next infrastructure‑breaking heatwave.
The missing sinks and small‑scale physics
One of the most basic blind spots lies in how models represent the land itself. Jan and colleagues point out that Some of those variables are missing from climate models entirely, including the way trees and land act as major sinks for carbon emissions. When I look at how forests respond to drought, fire, and shifting seasons, it is obvious that a static representation of “green cover” cannot capture the real‑time rise and fall of these sinks. Trees can flip from absorbing carbon to releasing it during die‑offs, yet many models still treat them as a steady sponge.
There is a similar problem with physical processes that occur at scales too small to resolve directly. Analysts have long warned that numerous inputs in climate models are not the direct result of scientific studies, but instead are approximations for things like cloud microphysics or turbulent eddies that are too small to model directly. When I connect that to the land‑surface gaps, the picture is of a system where key sinks and feedbacks are either smoothed over or guessed at. That does not invalidate the models, but it does mean their confidence bands are narrower on paper than they are in reality.
Why extremes keep outpacing the forecasts
The most visible sign that something is missing comes from extreme weather. Analysts have laid out at least five reasons why current climate models underestimate extreme events, and at the top of that list is a simple constraint: There is not enough data on the most violent storms, floods, and heatwaves to train or validate the simulations. By definition, extremes are rare, so the observational record is thin, especially outside North America and Europe. When I compare model output to the lived experience of communities in South Asia or sub‑Saharan Africa, the underestimation of compound disasters, like heat plus humidity or rain on snow, is stark.
Even within the physics‑based global models, researchers have flagged that, while the global climate models integrate many different physical parameters, they do not have everything. While the equations capture large‑scale circulation, they struggle with how individual thunderstorms organize, how heat domes lock in place, or how soil moisture feeds back into heatwaves. That is why I see models consistently underplaying the intensity and persistence of events like the Pacific Northwest heatwave or the multi‑year droughts that have hammered the Mediterranean and western United States.
Ocean shocks, Arctic signals, and the September surprise
Nowhere is the gap between model expectations and reality more vivid than in the oceans and polar regions. A recent analysis of the observed September 2023 temperature jump found that the month featured an unprecedented spike of nearly 0.6 °C above September 2022, a leap that standard climate models struggled to reproduce. The Abstract of that work notes that, although the long‑term warming trend is consistent with greenhouse forcing, the short‑term surge was amplified by water vapour feedback and other factors that are not always captured in coarse‑resolution simulations. When I look at that number, it reads less like a blip and more like a stress test the models were not fully prepared to pass.
Large‑scale patterns like the El Niño–Southern Oscillation add another layer of complexity. Forecasters at the Climate Prediction Center have warned that Even after equatorial Pacific sea surface temperatures in the Pacific transition to ENSO‑neutral, La Niña may still have some lingering influence through the atmosphere. That lag means models that key too heavily on surface temperatures can miss how a fading El Niño or emerging La Ni pattern will shape rainfall and storm tracks months later. When I connect that to the Arctic, where sea ice loss and stratospheric shifts are altering the jet stream, it becomes clear that the coupled ocean‑ice‑atmosphere system is evolving faster than the models’ parameterizations.
From La Niña winters to AI forecasts: patching the data gap
The stakes of these blind spots are immediate for seasonal forecasts. In the current outlook for the 2025–2026 winter, forecasters have highlighted how a sudden, dramatic shift in temperatures high in the atmosphere will dance with the La Niña weather pattern. La Niña, interacting with the polar vortex, is expected to shape where the deepest cold and heaviest snow set up, yet the range of possible outcomes remains wide. There will be communities that experience a winter far harsher than the ensemble average suggested, and that gap between expectation and reality is exactly where missing data on stratospheric waves and Arctic feedbacks bites hardest.
Researchers are turning to new tools to close that gap. Work highlighted by Jan on Arctic influences shows that While AI weather models have made impressive strides in short‑range forecasts of one to ten days, they still struggle to pick up the subtle Arctic signals that can foreshadow winter pattern shifts weeks in advance. At the same time, some scientists are probing other possible culprits for model‑observation mismatches, including glacial meltwater that can disrupt ocean circulation. When I put these strands together, the path forward looks less like ripping up the models and more like feeding them richer, more targeted data, from tree rings and deep‑ocean buoys to high‑resolution Arctic observations.
What “getting it wrong” really means for policy
It is tempting to treat every model miss as a failure of climate science, but the reality is more nuanced. The core equations of fluid dynamics and radiation that underpin these tools remain solid, and Jan is right to stress that they still capture the main trajectory of Earth’s warming. The real issue is that, in a world where infrastructure planners, farmers, and city mayors need guidance on specific extremes, a model that nails the global mean but misses the local shock is not enough. When I see a forecast that smooths over the risk of a once‑in‑a‑century flood that now arrives every decade, I read that as a data and resolution problem, not a reason to dismiss the entire enterprise.
For policymakers, the lesson is to treat model output as one line of evidence, not a precise script. That means stress‑testing plans against a wider envelope of possibilities, especially where the science itself flags missing pieces, from trees and land sinks to the evolving behavior of ENSO and the polar vortex. It also means investing in the observations that will let the next generation of models do better: denser sensor networks, long‑term ecological monitoring, and open data that lets AI systems learn from the full spectrum of extremes. Climate models got important details wrong because the world changed faster than the information we fed into them. The fix is not to abandon them, but to finally give them the data they have been missing.
More from Morning Overview