What is the Truth About Global Warming? The Real Meaning of "Climate Models are Wrong" — The Limits of the Precision Race Highlighted by Scientists

What is the Truth About Global Warming? The Real Meaning of "Climate Models are Wrong" — The Limits of the Precision Race Highlighted by Scientists

The provocative term "Rebellion Against Models"

"The Rebellion of Researchers Against Climate Models." Such a sensational headline does not indicate a collapse of the scientific consensus on anthropogenic global warming. Rather, it points to an internal scientific methodological debate about how climate models, long used to understand global warming, should be used, for what purposes, and with what accuracy.

Climate models calculate based on physical laws, including atmospheric and oceanic flows, radiation, clouds, sea ice, vegetation, and carbon cycles. Models developed by research institutions worldwide have been the foundation for recreating past climates, comparing scenarios with and without increased greenhouse gases, and considering future risks.

However, the Earth is not a laboratory apparatus. We cannot prepare two Earths and change the conditions, nor can we wait for the results of 2100 to grade prediction methods. Models are not replicas of reality but tools that extract parts of the complex Earth through equations and assumptions. This obvious fact becomes obscured in policy discussions and news headlines.


The central question is "Does more detail mean more accuracy?"

One key to understanding the current debate is a discussion by Hervé Douville of the French National Meteorological Research Center, published in PLOS Climate in July 2026. Douville, who served as a lead author for the IPCC Sixth Assessment Report's Working Group I, argued that while no climate model perfectly matches reality, they can be extremely useful if the purpose is correctly defined.

He particularly questioned the race towards ultra-high-resolution models that represent the Earth on a kilometer scale by refining the grid. Higher resolution could potentially allow for more direct representation of cumulonimbus clouds and ocean eddies, which were previously averaged. For municipalities wanting to understand local heavy rain or coastal changes, detailed geographical information is appealing.

However, "detailed" does not automatically mean "accurate." If errors remain in mechanisms like cloud-radiation interactions, carbon absorption by soil and forests, or long-term oceanic variations, simply refining the grid will not eliminate uncertainties. High-resolution modeling requires enormous computational resources, making it difficult to conduct numerous trials with varied initial conditions and parameters if focusing on a single precise calculation.

Douville's estimates suggest that transitioning from current-generation models to kilometer-scale models could typically require about 1,000 times the computational resources, including the burden of shortening time steps. The question is not about the performance of supercomputers per se, but about prioritizing whether to use limited resources for "a single ultra-precise future map" or to compare "many futures with varying coarseness."


Looking at a "bundle of futures" rather than a single future

There are broadly three uncertainties in the future of climate.

The first is the uncertainty of socio-economic scenarios regarding how much greenhouse gases will be emitted in the future. Energy policies, technological innovations, population, wars, economic conditions, and international cooperation cannot be predicted by physical laws alone. IPCC scenarios are not "prophecies" but comparative materials based on set conditions.

The second is the uncertainty of how to represent clouds, aerosols, oceans, and carbon cycles in models. Models from different research institutions, while based on common physical laws, differ in the representation and adjustment of detailed processes. Therefore, even with the same emission scenario, results can vary.

The third is the natural variability inherent in the climate system. Oceans and the atmosphere fluctuate chaotically, and even slight changes in initial conditions can make the trajectory of rain or temperature over decades in specific regions appear different. Even if the long-term global warming trend is the same, it may temporarily weaken in one region and become more pronounced in another.

For this reason, researchers use large-scale ensembles by slightly changing initial conditions rather than running a single model once. Douville points out that while five trials may suffice for some purposes, more than 100 may be necessary for others. In some cases, multiple results with variability are more useful for decisions on flood control, agriculture, water resources, and urban planning than a single detailed simulation.


"Climate sensitivity" is not a single definitive value

In discussions about climate models, equilibrium climate sensitivity, which indicates how much the Earth's average temperature will ultimately rise if carbon dioxide doubles, often appears. The IPCC Sixth Assessment Report estimates the best value at 3 degrees, with a likely range of 2.5 to 4 degrees, and a very likely range of 2 to 5 degrees.

It is a mistake to see this range and think "science knows nothing." The fact that carbon dioxide traps heat and human activity is the main cause of recent warming coexists with the range of final temperature responses, including feedback from clouds.

Furthermore, equilibrium climate sensitivity is not a forecast that "temperatures will definitely rise by 3 degrees by 2100." It is an indicator of the response when carbon dioxide concentration is fixed at double and the climate system, including the ocean, is allowed to equilibrate over the long term. The actual temperature this century will also be influenced by emissions, greenhouse gases other than carbon dioxide, air pollutants, and natural variability.


Admitting limitations is not a declaration of failure

The presence of bugs and adjustments in climate models can also easily heat up debates. A study on the ICON model developers, published in 2025, investigated how developers find, fix, and communicate bugs within massive codes that can exceed a million lines. Developers practically judge whether a model is sufficient to answer specific questions, not whether it is entirely error-free.

This is not a weakness of science in the same sense as aircraft design or drug efficacy assessment. It is the basic stance of model science to clarify the application range of models and verify whether the results withstand the purpose. The risk lies in users receiving detailed regional predictions as definitive futures without knowing those conditions.

On the other hand, past models were not entirely powerless. A study comparing global average temperature predictions published since the 1970s with later observations found that many models consistently captured subsequent warming when considering differences in actual radiative forcing. Models are not crystal balls predicting future weather by date and location, but they have a track record as tools for considering the long-term global response to increased greenhouse gases.

Therefore, the current criticism is not about "models being wrong, so climate action is unnecessary." Rather, it is a warning against promoting incomplete models as if they were perfect replicas of the Earth and leaving policy decisions to calculation results.


Why researchers are concerned about "colleagues' silence"

Model development has become a massive international project, linked with research funding, computational resources, international technological competition, and the climate services market. Higher resolution and flashy visualizations make it easier to explain the appeal of research plans. In contrast, explaining that "uncertainties remain despite precision" or that "running multiple existing models is more effective for certain purposes" is less glamorous.

Moreover, in politically charged environments surrounding climate change, there is caution that publicly discussing model weaknesses might be used to deny global warming. As a result, if scientists hesitate to voice legitimate internal criticism, it could damage trust in science when problems are discovered externally.

Discussing uncertainties openly does not weaken scientific consensus. Distinguishing "what is fairly certain, what still has a range, and in which applications confidence decreases" becomes the condition for trust.


Three reactions observed on social media

Reactions to public social media posts and article shares are not opinion polls and do not indicate the total volume or proportion of reactions. With that premise, the discussion is broadly divided into three directions.

The first is the reaction that takes the words "models are wrong" as a denial of all past global warming predictions. This overlaps with the question of whether it is right to proceed with costly decarbonization policies based on uncertain calculations. While this concern is important as a policy cost discussion, it is a leap to collectively deny the local and long-term limitations of models along with observations and physical laws indicating anthropogenic warming.

The second is the reaction that welcomes criticism as healthy scientific self-correction. This view favors disclosing conditions and error ranges and combining various models and observations over pretending precision. Posts by research-related individuals introducing the paper also raise questions about models sometimes underestimating recent changes and the need for more transparency in the relationship between physical processes and adjustments.

The third is the reaction that uncertainty is not a "reason to do nothing" but a "reason to prepare by assuming deviations." In investments like levees, drainage, heatstroke measures, agricultural varieties, and water source management, robust options are needed that consider not only a single central prediction but also deviations on the adverse side. This is like insurance, where not being able to predict the exact date of an accident does not render insurance meaningless.

On social media, the differences between specialized terms like "prediction," "projection," "scenario," and "ensemble" are often compressed into a single "hit/miss." If only sensational headlines are shared, contrary to researchers' claims, model criticism could become material for science denial.


What Japan needs is not "resolution," but the design of usage

In Japan, which faces disasters with significant regional differences such as typhoons, heavy rains, heatwaves, heavy snow, and droughts, there is a high demand for detailed climate information. Municipalities and companies must make decisions on factory and data center locations, river maintenance, crops, and power demand on specific regional units.

However, assuming that more detailed maps mean higher accuracy is dangerous. Even if color coding is displayed on a scale of a few kilometers, it does not necessarily mean that precipitation at a specific location decades ahead can be known with the same precision. Information providers need to show not only averages but also the range between models, natural variability, scenario dependency, and data limitations. Users should choose plans that minimize losses under multiple conditions rather than optimizing for a single figure.

Maintaining Japan's observation network will also become important. Douville pointed out that for the next IPCC assessment, the lack of continuous and managed observational data could be more of an obstacle than the limitations of existing Earth system models. Satellites, ocean buoys, weather radars, ground observations, and river/snow data are the foundation for evaluating, improving, and linking models to reality.


The right question is not "to believe or not"

Whether to believe in climate models or not—this binary choice is too crude a way to pose the problem. What should be asked is for what purpose the model was created, at what time and spatial scale it was verified, what assumptions it made, what it cannot represent, and how it shows the range of results.

Long-term global warming averages, regional precipitation decades ahead, next year's typhoon activity, and the necessary height of a specific levee are all different questions. They cannot be answered with the same model, resolution, or confidence level.

The maturity of climate science may not be about eventually completing a perfect Earth replica. It might be about measuring imperfections, comparing different models, constraining them with observations, and connecting them to robust decision-making even if deviations occur. This call for a shift in direction is now surfacing as a "rebellion of researchers."

Acknowledging the limitations of models makes science appear weak. But in reality, it's the opposite. Demonstrating what is unknown as unknown and explaining what can still be determined is the most solid path for science to regain trust amid the political noise surrounding climate.


Source URL

Related Articles