Scientists have known for decades that extreme heat is becoming more frequent. Both the research record and lived experience across the world document that shift. What has been far less understood is when these events occur within the year. Most research on seasonal warming has concentrated on average temperatures or on shifts in the timing of the seasons themselves. Summer conditions in mid-latitude regions, for example, have lengthened by roughly six days per decade since 1990. But the timing of individual extreme heat events—a different question entirely—had received little if any rigorous attention. Extremes may not move in step with the seasonal average, and that disconnection is precisely what the new study set out to measure.

The researchers expected that rising global temperatures would uniformly make it easier to cross dangerous heat thresholds throughout the year, widening the extreme heat season symmetrically at both ends. That is not what they found. Instead, the expansion is lopsided. “In some places, we have a larger expansion of extreme heat during the spring, before the traditional heat season starts. In other places, there’s a much faster expansion of the heat season into fall,” Ivanovich explains. The asymmetry means that different regions of the world are experiencing fundamentally different transformations of their heat regimes.

The methodology behind the findings was deliberately careful. The team counted extreme heat events on the six inhabited continents between 1980 and 1989, defining “extreme” as days falling in the hottest 5 percent of daily temperatures. They performed this counting twice, using two distinct measures of heat. The first was standard thermometer readings, a measure of dry heat. The second was wet bulb globe temperature, a more sophisticated metric that combines humidity, solar radiation and air temperature to quantify heat stress as the human body actually experiences it. This distinction matters enormously: humid heat limits the body’s ability to cool itself through sweating, making it considerably more dangerous to people, while dry heat is harder on crops and ecosystems. Adapting to one is not the same as adapting to the other.

The authors then compared those 1980s baseline figures to the most recent decade in the record, 2015 through 2024. The results were striking. Extreme heat seasons had expanded significantly across just over half of the world’s land area for dry heat, and just under half for humid heat. In the western United States, eastern China, northern Africa and eastern Europe, extreme heat events became more common more rapidly in the two months following their historical heat seasons. The opposite held in western Europe, southern Africa and northwestern India, where extreme heat arrived predominantly in the two months before the traditional season began. “There are very clear asymmetries in how extreme heat seasons are expanding in different parts of the world,” Ivanovich says. “Extreme heat is starting to become something different in a lot of these regions.”

To be certain the pattern was real and not an artifact of a single dataset, the researchers ran their analysis with two independent sources—one compiled by NASA and the other by the European Centre for Medium-Range Weather Forecasts. The pattern largely held up across both. This kind of replication is essential when studying extreme events, which are by definition rare. Extreme heat outside its season is rarer still, which makes changes in its timing statistically difficult to pin down. Ivanovich emphasizes that the findings should be taken as a compelling first line of evidence, and a next step is to repeat the comparison using climate simulations. Models can generate many more theoretical versions of reality under the same climatic conditions, providing a much larger sample of extreme events. If the models agree with what the observations show, that will strengthen the case that the observed changes represent a genuinely new pattern.

The timing of extreme heat matters for reasons that are both physiological and practical. Heat is harder on the human body when it arrives before people have acclimated to the season, or after they have already endured months of it. It is also harder for communities to prepare for, because cooling centers, heat alerts and public health campaigns are built around a summer calendar. A city that expects its heat season to end in September may find itself unprepared for a deadly hot spell in October. The data from Phoenix, Arizona, illustrate the phenomenon vividly. The city recorded 183 extreme heat days by temperature in the baseline decade of the 1980s, and 338 in the decade ending in 2024. None of the 1980s events fell after the heat season had ended, yet between 2015 and 2024, 6 percent did. The median date of the city’s dry heat extremes moved ten days later in the year, while its humid heat extremes moved 5.5 days earlier.

Phoenix’s recent experience underscores the stakes. After 113 consecutive days above 100 degrees Fahrenheit in 2024, stretching from late September into mid-October, the city went on to tie or break daily temperature records 21 days in a row. Maricopa County, where Phoenix is located, recorded 608 heat-related deaths in 2024, with 46 percent of them in July, the hottest month of the year. In 2023, July accounted for 64 percent of that year’s 645 deaths. And in March of this year, after the study period had ended, the city recorded nine days that topped 100 degrees Fahrenheit—something that had happened only once before in March over the entire historical record. The message is unambiguous: the shoulder seasons, once safe from dangerous heat, are no longer off-limits.

One of the most important questions the researchers addressed is what is driving the pattern. When they tested whether ordinary warming alone could reproduce it, rising average temperatures accounted for changes in the heart of the heat season but not for the lopsided expansion at its edges. Regional factors such as shifting rainfall patterns or changing land use are likely also at work. “We can’t confirm what share of the signal is due to climate change using observations alone,” Ivanovich says, but “it’s certainly the primary component of the story.” Climate models should help determine the respective contributions of human-induced climate change and natural variability, untangling how much of the asymmetry reflects a warming world and how much reflects local dynamics like soil moisture, irrigation and vegetation change.

The implications extend well beyond the heat season itself. Changes in the seasonal timing of extreme heat make it more likely that hot days will intersect with other seasonal hazards: peak wildfire season in the western United States, or peak hurricane season in the Southeast. When multiple hazards coincide or arrive in rapid succession, “they are much more dangerous and impactful than if these events happened in isolation,” Ivanovich says. A wildfire season that overlaps with an extended heat season strains emergency services, power grids and human health simultaneously. The study suggests that cities and public health agencies may need to rethink the entire architecture of heat preparedness, moving away from a summer-only framework toward one that treats dangerous heat as a year-round possibility in half the world’s land area. For the billions of people who live there, the hottest days of the year are no longer where the calendar says they should be.

News Publication Date: 10-Sep-2026

Web References: Not provided

References: Ivanovich, C., Cook, B., & McDermid, S. S. Dangerous Hot Days Are Spreading Beyond Summer. AGU Advances. https://www.eurekalert.org

Published in Communications Earth & Environment, the study by Ö. Kart Tokmak, L. Caesar, J. Ludescher and colleagues introduces an approach for transparent Earth system assessment. Its central idea is to combine the language-generation abilities of modern artificial intelligence with a retrieval system that searches relevant scientific sources before producing an answer. Instead of relying only on information encoded during a model’s training, the system can identify documents and evidence connected to a specific question, use them as context, and show users where its conclusions come from.

This architecture addresses one of the most serious weaknesses of generative AI: a system can produce fluent, confident text even when it is missing key evidence or has misunderstood a technical concept. In scientific applications, that failure is not merely embarrassing. An incorrect statement about climate risks, tipping points, carbon-cycle feedbacks, or adaptation strategies could influence public communication and policy decisions. Retrieval-augmented generation, commonly abbreviated as RAG, is intended to reduce this danger by grounding responses in selected external material rather than allowing the model to answer entirely from statistical patterns.

The framework is described as domain-adapted because general-purpose AI systems are not automatically prepared for the language and structure of Earth system research. Climate science, for example, depends heavily on specialized terminology, carefully defined variables, geographic references, time periods, model ensembles, and statements about uncertainty. The word “risk” can have a precise technical meaning in one assessment, while “sensitivity,” “forcing,” “resilience,” or “abrupt change” may refer to concepts that cannot be interpreted safely without their scientific context. A domain-adapted system is designed to recognize these distinctions and retrieve information accordingly.

A crucial feature of the proposed approach is transparency. In a conventional chatbot interaction, a user may receive a polished paragraph without knowing which evidence supports it, whether the sources disagree, or how current the information is. A retrieval-based framework can expose the documents, passages, or references used to construct an answer. That creates an evidence trail, allowing researchers to check whether a statement accurately reflects the cited material. It can also reveal when a question is poorly supported, when sources conflict, or when the available literature is too limited for a confident conclusion.

The system’s relevance extends beyond producing summaries. Earth system assessment often requires connecting findings across disciplines: atmospheric physics, oceanography, hydrology, ecology, economics, and social science may all contribute to a single question. These fields do not always use identical terminology or organize knowledge in the same way. A retrieval framework can help assemble related evidence across those boundaries, while domain-specific instructions and source selection can limit the risk that an apparently relevant passage is technically unsuitable. The goal is not to replace expert judgment, but to make the process of locating and comparing evidence faster and more systematic.

For scientists, such a tool could become a research companion capable of scanning large collections of papers and assessment documents, identifying relevant passages, and organizing information around a defined question. For decision-makers, it could provide a more accessible route into highly technical literature—provided that every answer remains tied to verifiable sources. For communicators and educators, the ability to distinguish established findings from uncertain or contested claims could be especially valuable at a time when climate information is frequently simplified, distorted, or stripped of its qualifications.

The framework also highlights why artificial intelligence used in science must be evaluated differently from ordinary consumer applications. Fluency is not enough. A useful Earth system system must retrieve appropriate sources, preserve the meaning of technical language, respect publication dates, distinguish observations from projections, and communicate uncertainty without burying it. It must also resist the temptation to produce a single definitive answer when the scientific literature contains multiple plausible interpretations. Transparency therefore becomes a technical requirement, not just a desirable feature: users need to understand both what the system knows and how it reached its response.

The study arrives as institutions around the world search for ways to turn the expanding climate knowledge base into practical guidance. The amount of information will continue to grow as new satellite missions, high-resolution models, sensor networks, and scientific publications generate evidence at unprecedented speed. A domain-adapted retrieval-augmented framework could help transform that flood of material into structured, traceable assessments. Its real test, however, will be whether experts can reproduce its answers, identify its limitations, and trust it without surrendering responsibility for scientific interpretation. If those conditions are met, AI may become less a mysterious answer machine and more a transparent research instrument for understanding a rapidly changing planet.

Subject of Research: Domain-adapted retrieval-augmented artificial intelligence for transparent Earth system assessment.

Article Title: A domain-adapted retrieval-augmented framework for transparent Earth system assessment.

Article References: Kart Tokmak, Ö., Caesar, L., Ludescher, J. et al. “A domain-adapted retrieval-augmented framework for transparent Earth system assessment.” Communications Earth & Environment 7, 648 (2026). https://doi.org/10.1038/s43247-026-03878-1

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s43247-026-03878-1

Keywords: Earth system science, climate assessment, artificial intelligence, retrieval-augmented generation, domain adaptation, scientific transparency, evidence retrieval, climate information.

The researchers meticulously analyzed climate data from a variety of sources to gauge the extent and significance of the temperature anomaly. Their findings revealed an unprecedented spike in average temperatures across multiple regions, raising questions about the stability of climate models and our understanding of anthropogenic influences. The team’s work emphasizes not only the necessity of scrutinizing existing models but also the importance of understanding the potential for extreme events that could deviate from typical patterns observed in recent decades.

The study underscores the notion that extreme weather events may become increasingly common as climate change continues to evolve. Traditional models of anthropogenic forcing generally predict a gradual increase in temperatures correlated with rising greenhouse gas emissions. However, the September 2023 temperature jump appears to defy these expectations, suggesting that additional factors may be influencing global temperatures in unexpected ways. This highlights a crucial gap in current climate science that warrants further exploration.

One key aspect of the research focuses on the interplay between natural climate variability and anthropogenic influences. While human activities have been shown to significantly impact global temperatures, the researchers assert that natural climate phenomena—such as ocean currents and solar radiation—can still exert formidable effects on weather patterns and temperature fluctuations. The interaction between these natural and anthropogenic factors might well contribute to the emergence of extreme temperature anomalies.

In addition to exploring the potential for natural variabilities, the authors also evaluate the limitations of existing climate models in capturing the complexities of climate systems. They argue that many models may oversimplify the interactions within the Earth’s climate and atmospheric systems, potentially leading to a significant underestimation of the likelihood of extreme temperature variations. As such, there is an urgent need for refining these models to accommodate a broader range of climatic interactions.

The research team utilized a combination of satellite data, ground-based measurements, and climate simulations to analyze the phenomena surrounding the temperature spike. Their comprehensive methodology allowed them to cross-reference findings and validate their conclusions while ensuring robustness in their analyses. The data collected paints a vivid picture of an Earth experiencing unusual climatic shifts, thus enhancing the urgency to reassess our current understanding of climate change.

Importantly, this study provokes a discussion regarding the broader implications of the September 2023 temperature jump on policy decisions related to climate action. If extreme temperature events are becoming more frequent and severe, as the authors suggest, it becomes increasingly critical for policymakers to prioritize adaptive strategies and mitigation measures. Understanding the dynamics of such extreme events could inform regulations and initiatives that aim to curb emissions and enhance climate resilience.

The findings of Seeber and colleagues underscore the need for a paradigmatic shift in how climate science is approached. As climate change unfolds, it represents an evolving challenge that necessitates both innovative research methodologies and a willingness to question long-held assumptions. Scientists must remain open to the idea that our understanding of climate dynamics is not fixed; rather, it is dynamic and likely to change as new data emerges.

Furthermore, the implications of the September 2023 anomaly extend beyond scholarly pursuits and into the realm of public understanding. Given the pervasive impacts of climate change, it is essential that the broader community and stakeholders recognize the potential for abrupt changes. Disseminating this knowledge is vital for fostering public discourse and ensuring that communities are prepared for possible future climatic extremes.

As this research takes center stage within the scientific community, it may also capture the attention of the general population. There is a growing desire for insights that bridge the gap between complex science and everyday experiences. The narrative surrounding this temperature jump might resonate well with those seeking to comprehend the nuances of climate change and its real-time implications.

In conclusion, the intriguing findings presented by Seeber et al. serve as a call to action for researchers, policymakers, and the general community alike. The September 2023 temperature anomaly challenges existing paradigms and demonstrates the critical need for a holistic approach to understanding climate variability. It serves as a reminder that the science of climate change is continuously evolving and that staying informed about these developments is essential for navigating the future.

Through enhanced communication and collaboration among the scientific community, industry, and governments, it is possible to develop comprehensive strategies that stem from these newfound insights. The findings of this study, while alarming, also provide a unique opportunity to rethink our approaches to climate adaptation and resilience in the face of unprecedented environmental change.

Subject of Research: Temperature anomalies and anthropogenic climate forcing

Article Title: The observed September 2023 temperature jump was nearly impossible under standard anthropogenic forcing.

Article References:

Seeber, S., Schumacher, D.L., Gudmundsson, L. et al. The observed September 2023 temperature jump was nearly impossible under standard anthropogenic forcing.
Commun Earth Environ (2026). https://doi.org/10.1038/s43247-026-03178-8

Image Credits: AI Generated

DOI: 10.1038/s43247-026-03178-8

Keywords: Climate change, temperature anomaly, climate models, anthropogenic forcing, extreme weather events.