At the heart of this investigation lies a direct comparison between traditional physics-based climate emulators and state-of-the-art deep-learning models. Climate emulators serve as streamlined proxies for comprehensive climate models that otherwise demand prodigious computational resources and extended runtimes on supercomputers. These emulators are critical for policymakers who require near-real-time assessments of potential future climate scenarios based on varying greenhouse gas emissions. The MIT researchers embarked on an analytical journey to assess the fidelity of these disparate modeling approaches under controlled experimental settings.

The study’s initial findings surprisingly indicated that a traditional method known as linear pattern scaling (LPS) consistently surpassed deep-learning models in forecasting a broad spectrum of climate parameters. LPS, rooted in the fundamental physics governing climate systems, leverages linear relationships to extrapolate changes in environmental factors in response to external forcings such as pollution or greenhouse gas concentration. This approach contrasts with deep-learning models, which seek to infer patterns directly from raw data without explicit physical constraints. The advantage exhibited by LPS across multiple parameters, including critical variables like temperature and precipitation, was unexpected given the non-linearity inherent in many climate processes.

Further scrutiny revealed that the apparent superiority of LPS in many benchmarks was, in part, a consequence of natural internal climate variability that deep-learning models struggled to capture accurately. Climate systems exhibit intrinsic oscillations, such as the El Niño-Southern Oscillation, which introduce significant fluctuations in meteorological phenomena over multi-year timescales. The authors identified that these long-term oscillations confounded the deep-learning models, leading to poorer predictions in the presence of high natural variability. LPS, with its smoothing effect, tended to average out these oscillations, artificially enhancing its apparent prediction accuracy during benchmark evaluations.

Recognizing the limitations imposed by conventional evaluation metrics, the research team devised a more rigorous benchmarking methodology that explicitly accounts for natural variability. This novel evaluative framework employs extensive datasets encompassing multiple climate model runs, thereby isolating the impacts of internal variability from the model’s predictive skill. Under this recalibrated benchmark, deep-learning models demonstrated a marked improvement in predicting local precipitation patterns, a notoriously challenging task due to its spatial heterogeneity and episodic nature. Yet, for regional surface temperature projections, LPS still maintained a slight edge, underscoring the nuanced performance differences between the two approaches under varying climatic variables.

The implications of these findings resonate deeply within the climate modeling community. While deep-learning techniques offer unparalleled flexibility and the potential to model highly nonlinear processes, their current formulations may not yet fully capitalize on domain-specific physical knowledge embedded in classical models. This mismatch suggests that future advancements in climate machine learning will require hybrid approaches that integrate physical laws with data-driven methods to achieve superior predictive accuracy, especially at finer spatial and temporal resolutions.

Enhancing the practical utility of these insights, the MIT team incorporated the LPS method into a climate emulation platform designed to provide rapid assessments of local temperature responses under assorted emissions scenarios. Such platforms are indispensable for informing policymakers who must weigh the economic and social ramifications of regulatory decisions against anticipated climatic outcomes. By ensuring the underlying emulator relies on the most reliable mathematical formalism, this work bolsters confidence in decision-support tools essential to global climate governance.

However, the researchers caution against viewing LPS as a panacea. While it offers robustness in capturing mean trends, LPS lacks the capacity to simulate variability and extreme weather phenomena, which are increasingly salient in climate impact assessments. Deep-learning models, with their capacity to account for complex nonlinear dynamics, hold promise in addressing these deficiencies, especially when developed alongside more sophisticated evaluation frameworks.

A crucial takeaway from this study is the paramount importance of establishing robust benchmarking standards that transparently appraise model performance in the context of intrinsic climate variability. Without such standards, comparative assessments risk misconstrued conclusions, potentially privileging models ill-suited for real-world deployment. The researchers advocate for an expansion of benchmarks to include more impact-focused metrics, such as drought severity indices or wildfire occurrence probabilities, to better align model evaluation with decision-maker priorities.

The endeavor also illuminates fertile grounds for further research. Future work could explore the synergy between machine-learning models and physically grounded emulators, utilizing advances in adaptive systems and computational mathematics to reconcile the strengths of each. Additionally, probing under-explored climatic variables, such as regional wind circulations or aerosol interactions, may unlock new frontiers in predictive fidelity.

Ultimately, this study underscores a humble yet critical principle: bigger and more complex does not invariably equate to better in climate modeling. Thoughtful integration of physics-based insights, paired with judicious use of machine learning, holds the key to delivering models that are not only scientifically robust but also operationally meaningful. As the urgency of climate action escalates, equipping policymakers with trustworthy, actionable predictions remains an overarching priority—one that demands continual refinement of both models and the metrics by which they are judged.

In summary, the MIT-led research offers a sobering but hopeful perspective on the evolving role of AI in climate science. It highlights the dangers of uncritical adoption of large AI models detached from domain expertise and champions the development of hybrid approaches grounded in physical understanding. With its novel benchmarking framework and nuanced performance analysis, the study charts a course towards more reliable, interpretable climate emulators that can better serve humanity’s quest to navigate an uncertain climatic future.

Subject of Research: Computational simulation and modeling of climate prediction methods, focusing on benchmarking deep learning and physics-based climate emulators.

Article Title: The Impact of Internal Variability on Benchmarking Deep Learning Climate Emulators

News Publication Date: August 26, 2025

Web References:
https://climategrandchallenges.mit.edu/flagship-projects/bringing-computation-to-the-climate-challenge/
https://bc3.mit.edu/demos/en-roads/
http://dx.doi.org/10.1029/2024MS004619

References:
Lütjens, B., Selin, N., Ferrari, R., Watson-Parris, D. (2025). The Impact of Internal Variability on Benchmarking Deep Learning Climate Emulators. Journal of Advances in Modeling Earth Systems. DOI: 10.1029/2024MS004619

Keywords: Artificial intelligence, climate change, machine learning, computer modeling, computational simulation, climate emulators, internal variability, deep learning, linear pattern scaling

At the core of this research lies the examination of rolling 31-year latitude–year Hovmuller plots capturing zonal mean temperature anomalies at the 200 hPa pressure level—a key altitude for upper tropospheric circulation. The focus on hemispheric temperature gradients aligns with classical atmospheric dynamics theory, whereby meridional temperature differentials directly influence jet stream position and intensity. Notably, studies illustrate how transient volcanic eruptions in the twentieth century, such as those of Krakatau (1883), Agung (1963), El Chichon (1982), and Pinatubo (1991), induce marked tropical cooling. This cooling phenomenon systematically weakens the meridional temperature gradient, quantified as (\frac{\partial \overline{T}}{\partial \phi}), triggering an equatorward displacement of the jet stream that correlates with a tendency toward negative NAO phases.

Conversely, the absence of volcanic aerosol injections in the early twentieth and twenty-first centuries coincides with sustained tropical warming, amplifying the meridional temperature gradient and thereby pushing jet streams poleward. This dynamic fosters a predisposition toward positive NAO phases, altering pressure distributions that affect storm tracks, precipitation patterns, and temperature regimes across North Atlantic bordering continents. Intriguingly, the peak cooling effect within the hist-nat (historical natural-only forcings) simulations centers around 1990, attributed to the cumulative impact of overlapping volcanic events such as Agung, El Chichon, and Pinatubo. This timing contrasts with the historical simulations incorporating anthropogenic GHGs, where the minimum in tropical temperature and corresponding NAO shift occurs earlier, underscoring the complex temporal interplay between natural and anthropogenic influences.

Investigations extend to future climate scenarios, modeled under representative concentration pathways SSP1-2.6, SSP2-4.5, and SSP5-8.5, projecting divergent tropical temperature trajectories. Under the high-emission SSP5-8.5, tropical warming intensifies unabated, reinforcing meridional gradients and potentially driving NAO magnitudes to levels unprecedented in the paleoclimate record. Mitigation-aligned pathways SSP2-4.5 and particularly SSP1-2.6 illustrate decelerated warming trends or even tropical temperature declines, hinting at opportunities to attenuate such extreme oscillatory behavior. These projections exemplify the profound impact of emission choices on upper atmospheric circulation and midlatitude climate variability.

The research rigorously quantifies trends in the meridional temperature gradient at 35°N, a latitude central to Northern Hemisphere midlatitude jet positioning, showing that its variability serves as a robust explanatory variable for past and projected NAO fluctuations. Rolling 31-year averages of (-\frac{\partial \overline{T}}{\partial \phi}) align closely with calibrated NAO indices derived from hist-nat forcings in the reanalysis and climate model ensembles. The negative sign accounts for the generally negative gradient in the Northern Hemisphere, ensuring that increases in gradient magnitude correspond coherently with NAO strength.

While this temperature gradient mechanism cogently explains many observed features of NAO variability, the authors acknowledge that additional atmospheric and oceanic processes inevitably contribute. Prior research suggests influences from regional sea ice changes, remote tropical oceanic variability—especially in the Indian Ocean—and complex modes such as the Northern Annular Mode. Such multifactorial interactions demand holistic climate modeling and observational studies to fully unravel mechanisms underpinning multidecadal NAO changes.

The implications of these findings are manifold. The NAO modulates precipitation regimes critical for agriculture, water resources, and ecosystems across North America and Europe. Dramatic alterations in NAO magnitude and phase can exacerbate droughts, floods, and temperature extremes, compromising societal resilience. Hence, understanding the drivers of NAO variability offers critical foresight for climate adaptation and risk management frameworks.

Central to the forecasted risk is the relationship between midtropospheric tropical temperature anomalies and the broader circulation patterns. The cooling induced by volcanic aerosols injects shortwave attenuation and a resultant radiative forcing that temporarily depresses tropical temperatures, weakening the zonal mean temperature gradient. This gradient influences the subtropical jet stream’s latitudinal position, which in turn modulates the strength and phase of NAO. The detailed latitude-year Hovmuller analyses provide a climate diagnosis tool visualizing temporal and spatial evolution of these processes, revealing coherent shifts linked to specific volcanic and GHG forcing scenarios.

Moreover, the models capture intrinsic feedback loops whereby greenhouse gas warming alters stratospheric temperature and circulation, further modifying tropospheric temperature gradients. The persistence of these changes at 200 hPa illustrates the vertical coupling between stratosphere and troposphere, critical to capturing realistic NAO dynamics. The study’s multi-model ensembles ensure robustness against model biases, enhancing confidence in projected outcomes.

Encouragingly, scenarios with aggressive mitigation targeting lower emissions demonstrate a potential dampening of NAO variability. This suggests that emission pathways are not merely drivers of global mean temperature but also crucial moderators of extremal atmospheric circulation patterns. The research thereby informs policy discourse by linking terrestrial emission targets with regional climate stability.

Despite these advances, the complexity of NAO modulation demands continued exploration. Factors such as sea surface temperature anomalies in the tropical Indian Ocean, Arctic sea ice variability, and teleconnections with other atmospheric modes remain areas for intensive study. Understanding their relative contributions alongside the meridional temperature gradient framework remains a frontier in climate dynamics.

Beyond the scientific contribution, this research provides granular insights that can be integrated into regional forecasting tools and climate risk assessments. The capacity to anticipate shifts in the NAO under anthropogenic forcing scenarios enables governments, insurers, and urban planners to better prepare for extreme weather events and hydrological variability.

In summary, the new findings emphasize a fundamental linkage between tropical upper-tropospheric temperature gradients and the multi-decadal behavior of the North Atlantic Oscillation, shaped by the interplay of volcanic forcing and anthropogenic greenhouse gas emissions. Mitigation strategies aimed at limiting warming prove essential not only for global temperature stabilization but also for preserving stable and predictable atmospheric circulation patterns that underpin midlatitude climate regimes.

As nations confront the challenges of climate change, elucidating and acting on the drivers of critical oscillations such as the NAO represents a pressing research and policy priority. This study provides a compelling case for the integration of comprehensive mitigation efforts to avert unprecedented circulation anomalies with far-reaching climatic and societal consequences.

Subject of Research: North Atlantic Oscillation variability and its modulation by tropical upper-atmospheric temperature gradients under natural and anthropogenic forcings.

Article Title: Mitigation needed to avoid unprecedented multi-decadal North Atlantic Oscillation magnitude.

Article References:
Smith, D.M., Dunstone, N.J., Eade, R. et al. Mitigation needed to avoid unprecedented multi-decadal North Atlantic Oscillation magnitude. Nat. Clim. Chang. 15, 403–410 (2025). https://doi.org/10.1038/s41558-025-02277-2

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41558-025-02277-2