While the trajectory of global warming is relatively well understood, changes in climate variability and associated extremes remain uncertain. Understanding these changes is crucial because increased variability poses a greater risk to both ecosystems and human society than changes in mean temperature alone. Previous research using instrumental records, single climate model simulations, or multi-model ensembles has yielded inconclusive results regarding projected changes in temperature variability, with estimates ranging from no change to regional increases. This study leverages the unique capabilities of single-model initial-condition large ensembles (SMILEs) from multiple models to disentangle the forced response from internal variability, enabling a more robust assessment of projected changes in temperature variability. The study further contextualizes these projections with evidence from instrumental records, paleoclimate proxies, and past climate simulations, such as the Last Millennium Ensemble of the Community Earth System Model (CESM1-CAM5 LME), aiming to provide a more comprehensive understanding of the future evolution of temperature variability.

The authors review previous studies on temperature variability, highlighting the inconsistencies in findings from traditional methods like instrumental records, single model simulations, and multi-model ensembles from CMIPs. These studies have produced conflicting results, ranging from no change, slight global decreases, to regional increases in temperature variability. The challenges in disentangling the forced response and internal variability using traditional tools have resulted in inconclusive estimates of future changes. This study addresses the limitations of previous studies by using SMILEs, which allow the separation of internal variability from externally forced variability, providing more reliable estimates of projected changes.

To quantify the evolution of regional near-surface air temperature variability from 850 to 2100 CE, the study combines data from instrumental records, proxy reconstructions, and several model simulations. Data and simulations are interpolated to a 1° × 1° spatial resolution. The forced response is removed from the SMILEs to isolate internal variability. The study compares the global average of local standard deviations from observations and model simulations, finding agreement in magnitude and relative stability since 850 CE. The magnitude of variability in global mean temperature is assessed by comparing natural variability from the PAGES2k database with the CESM1-CAM5 LME. Spatial distribution of temperature variability for the last millennium is evaluated by comparing the N-TREND database with PMIP3 and CESM1-CAM5 LME. To compare SMILE-simulated changes with observed changes, the similarity between instantaneous ensemble variability and temporal standard deviation after detrending is exploited. The forced signal is removed by detrending instrumental records with the multi-model mean of SMILE means. The standard deviation ratio between 1970–2019 and 1920–1969 is calculated for both SMILEs and observational products. To assess the patterns of regional variability change, the study compares the patterns of change with the magnitude of variability caused by natural external forcings using the CESM1-CAM5 LME. Emergence of anthropogenically forced change is determined by comparing the future temperature variability with the range of past unforced temperature variability derived from preindustrial control simulations. The study also investigates the underlying mechanisms of the latitudinal contrasting patterns of temperature variability change, focusing on sea ice loss at high latitudes and changes in vegetation cover in the tropics. The Bowen ratio is used as a proxy for vegetation cover, and its relationship with changes in temperature variability over land is analyzed.

The study finds that anthropogenically forced changes in internal interannual temperature variability are projected to emerge from the unforced range of internal variability over the 21st century. While globally averaged regional temperature variability decreases only slightly across the model average, a contrasting pattern of increasing variability over tropical land and decreasing variability in high latitudes is projected under strong global warming. The observed pattern of temperature variability change is consistent with the simulated pattern, showing increased variability on tropical and subtropical land and the central and eastern Pacific, and decreased variability at mid-latitudes. While models show inconsistent globally averaged change, they consistently project substantial decreases in temperature variability at high latitudes and increases at low latitudes, particularly over land. The decrease in high-latitude variability is primarily attributed to sea ice loss, while the increase in low-latitude variability is linked to a transition to drier surface types. The study finds that the anthropogenically forced changes in tropical land temperature variability are likely to exceed naturally forced changes, making human influence a major driver of historic temperature variability change. The emergence of anthropogenically forced change from unforced variability is robust, with most SMILEs agreeing on broad features of emergence at the end of the 21st century. Hotspot regions of increased temperature variability include the Amazon, Southeast Asia, Australia, and West Africa.

The findings of this study address the research question by demonstrating that human-caused climate change will lead to significant and widespread changes in year-to-year temperature variability. The distinct pattern of regional changes—increased variability in the tropics and decreased variability at high latitudes—is a key contribution, exceeding the range of natural variability observed over the past millennium. The mechanisms identified—sea ice loss and changes in vegetation—provide a plausible explanation for the observed and projected changes. The study's significance lies in its robust quantification of these changes using SMILEs, providing stronger evidence than previous studies based on limited datasets or single model outputs. The results highlight the need for urgent mitigation efforts to avoid the projected unprecedented changes in temperature variability and associated climate extremes.

This study provides robust evidence for the emergence of large-scale, human-caused changes in year-to-year temperature variability by the end of the 21st century. The contrasting patterns of increased variability in tropical regions and decreased variability in high latitudes are supported by multiple lines of evidence, including climate model simulations, observations, and paleoclimate reconstructions. The identified mechanisms, sea ice loss and changes in vegetation cover, explain the observed patterns. These findings highlight the urgent need for substantial emission reductions to avoid unprecedented changes in temperature variability and associated extreme weather events. Future research could focus on improving the representation of regional climate processes in climate models to refine projections of temperature variability change and improve understanding of the complex interactions between vegetation, soil moisture, and temperature.

The study acknowledges limitations stemming from data gaps in instrumental records, particularly in key regions such as tropical land and high-latitude oceans, which limit the confidence in the observed pattern. The differences in the representation of vegetation dynamics and sea ice processes across the various climate models could also contribute to uncertainties in the simulated temperature variability changes. The study relied on a specific set of climate models, and the results might differ if other models were included.