Driven by growing impacts of changing precipitation amounts and phase on the Arctic's natural and built environment, we examine seasonal patterns and trends in Arctic precipitation and partitioning between its liquid and solid forms. Use is made of data from ERA5 reanalysis, Automated Surface Observing System stations over land, and a climatology based on present weather reports over the Arctic Ocean. In the Atlantic sector of the Arctic, most precipitation falls in liquid form in all seasons in its southern limits, but snowfall is high over its northern parts. Precipitation over the dry central Arctic Ocean and terrestrial polar deserts almost always falls as snow. Even during the summer, typically 50 % of precipitation over the central Arctic Ocean falls as snow. Over land, nearly all summer precipitation falls in liquid form, except in the Canadian Arctic Archipelago where summer snowfall is still common. Annual precipitation has increased since 1979, primarily in the Barents Sea sector, accompanied by generally downward trends in snowfall and, hence, upward trends in liquid precipitation. Across much of the Arctic, the liquid to total precipitation ratio has increased only in summer and autumn, while in the Atlantic sector, the liquid to total precipitation ratio has increased in fall and winter.

Sea-ice stage of development (SoD) describes the age and associated thickness of sea ice and provides important information for navigation, and operational ice monitoring. SoD labels are obtained from operational ice charts, where trained analysts interpret satellite observations and assign standardized stage codes to regions with similar ice conditions. These codes often represent ranges of compatible ice thicknesses rather than exact physical values. Deep-learning methods can automate SoD mapping and commonly adopt operational ice charts as reference labels for training. These annotations are not exact, however; this is because chart interpretation relies on analyst judgement and on the observations available at the time, so different ice services may assign different SoD labels to the same conditions. We term this variation across independently produced expert annotations multi-annotator label uncertainty; collapsing the annotations into a single deterministic target discards this variation. A second source of uncertainty originates in the learned model itself. In this paper, we quantify both sources: annotation uncertainty from disagreement among independent ice-service charts and model uncertainty from the learned predictive models. We then evaluate their relationship by testing whether model uncertainty is higher where ice services disagree. We observe that supervision incorporating information from multiple annotators can improve this correspondence, with soft supervision achieving the highest overall correlation of 0.256. The relationship becomes substantially stronger near the ice edge, where model predictive uncertainty closely tracks multi-annotator disagreement, reaching a correlation of 0.704 within 0–10 km. Among the uncertainty-estimation approaches, Monte Carlo dropout provides the best-calibrated confidence estimates, with an expected calibration error of 0.050.

Light transmission through a sea ice cover has strong implications for the heat content of the upper ocean, the magnitude of bottom and lateral ice melt, and primary productivity in the ocean. Light transmittance in the vicinity of the Multidisciplinary Drifting Observatory for the Study of Arctic Climate (MOSAiC) Central Observatory was estimated by driving a two-stream radiative transfer model with physical property observations. Data include point and transect observations of snow depth, surface scattering layer thickness, ice thickness, and pond depth. The temporal evolution of light transmittance at specific sites and the spatial variability along transect lines were computed. Ponds transmitted 4–6 times as much solar energy per unit area as bare ice. On July 25, ponds covered about 18% of the area and contributed roughly 50% of the sunlight transmitted through the ice cover. Approximating the transmittance along a transect line using average values for the physical properties will always result in lower light transmittance than finding the average light transmittance using the full distribution of points. Transmitted solar energy calculated using the standard five ice thickness categories and three surface types used in the Los Alamos sea ice model CICE, the sea ice component of many weather and climate models, was only about 1 W m−2 less than using all the points along the transect. This minor difference suggests that the important processes and resulting feedbacks relating to solar transmittance can be represented in models that use five or more categories of ice thickness distributions.

Sea ice plays a critical role in the global climate system and maritime operations, making timely and accurate classification essential. However, traditional manual methods are time-consuming, costly, and have inherent biases. Automating sea-ice type classification addresses these challenges by enabling faster, more consistent, and scalable analysis. While both traditional and deep-learning approaches have been explored, deep-learning models offer a promising direction for improving efficiency and consistency in sea-ice classification. However, the absence of a standardized benchmark and comparative study prevents a clear consensus on the best-performing models. To bridge this gap, we introduce IceBench, a comprehensive benchmarking framework for sea-ice type classification. Our key contributions are three-fold: First, we establish the IceBench benchmarking framework, which leverages the existing AI4Arctic Sea Ice Challenge Dataset as a standardized dataset, incorporates a comprehensive set of evaluation metrics, and includes representative models from the entire spectrum of sea-ice type-classification methods categorized in two distinct groups, namely pixel-based classification methods and patch-based classification methods. IceBench is open-source and allows for convenient integration and evaluation of other sea-ice type-classification methods, hence facilitating comparative evaluation of new methods and improving reproducibility in the field. Second, we conduct an in-depth comparative study on representative models to assess their strengths and limitations, providing insights for both practitioners and researchers. Third, we leverage IceBench for systematic experiments addressing key research questions on model transferability across seasons (time) and locations (space), data downsampling, and preprocessing strategies. By identifying the best-performing models under different conditions, IceBench serves as a valuable reference for future research and a robust benchmarking framework for the field.

Capturing spatial context at multiple scales is crucial for deep learning-based sea ice segmentation. However, the optimal specification of spatial context based on observation resolution and task characteristics remains underexplored. This study investigates the impact of spatial context on the segmentation of sea ice concentration, stage of development, and floe size using a multi-task segmentation model. We implement Atrous Spatial Pyramid Pooling with varying atrous rates to systematically control the receptive field size of convolutional operations, and to capture multi-scale contextual information. We explore the interactions between spatial context and feature resolution for different sea ice properties and examine how spatial context influences segmentation performance across different input feature combinations from Sentinel-1 SAR and Advanced Microwave Radiometer-2 (AMSR2) for multi-task mapping. Using Gradient-weighted Class Activation Mapping, we visualize how atrous rates influence model decisions. Our findings indicate that smaller receptive fields excel for high-resolution Sentinel-1 data, while medium receptive fields yield better performances for stage of development segmentation and larger receptive fields often lead to diminished performances. The fusion of SAR and AMSR2 enhances segmentation across all tasks. We highlight the value of lower-resolution 18.7 and 36.5 GHz AMSR2 channels in sea ice mapping. These findings highlight the importance of selecting appropriate spatial context based on observation resolution and target properties in sea ice mapping. By systematically analyzing receptive field effects in a multi-task setting, our study provides insights for optimizing deep learning models in geospatial applications.

Accurate segmentation and mapping of sea ice types is crucial for safe polar navigation, offshore operations, climate monitoring, and ecosystem analysis. While deep learning has demonstrated strong potential for automating sea ice type segmentation, its success often relies on access to extensive expert-labeled datasets, which is both resource-intensive and time-consuming to create. However, foundation models (FMs), recently developed through self-supervised training on large-scale datasets, have demonstrated impressive performance across a range of remote sensing downstream tasks. Nevertheless, their applicability to sea ice type segmentation based on Synthetic Aperture Radar (SAR) imagery remains uncertain due to the unique challenges posed by sea ice-such as intricate geophysical patterns, pronounced seasonal variability, and SAR-specific artifacts like banding, scalloping, and heterogeneous backscatter-as well as the fact that SAR data in polar regions are often acquired using specialized sensor modes that differ markedly from those used to collect FM training data at lower latitudes, limiting their direct transferability to polar environments. To address this gap, we contribute: (1) Ice-FMBench, a comprehensive benchmark framework for evaluation of the state-of-the-art remote sensing FMs on the sea ice type segmentation task using Sentinel-1 SAR imagery, where Ice-FMBench is composed of a widely used standardized dataset, diverse evaluation metrics, and a representative set of selected remote sensing FM models most suitable for sea ice type segmentation, with the ability to include new models side-by-side the existing models; (2) an extensive comparative evaluation of the representative FMs using Ice-FMBench, with additional case studies to assess performance of the top-performing model in terms of transferability across temporal and spatial domains and sensitivity to training dataset size; and (3) a multi-teacher knowledge distillation approach to address lack of spatiotemporal transferability of the existing FMs by transferring insight from spatially and temporally specialized expert models into a single, efficient student model.

Sea ice type classification is essential for climate change research and maritime safety. Traditionally, this process relies on manual ice charting, which is time-consuming, expensive, and requires expert knowledge, making it difficult to scale up for current demands. Automating sea ice type classification is essential to keep pace with rapidly changing sea ice conditions. However, two main challenges limit the development of effective automated classifiers. First, while ice charts provide valuable labeled data, they only offer large-area (polygon) annotations rather than pixel-level labels, leading to a lack of precise training data. Second, although there are additional datasets with useful sea ice information, effectively combining these different data sources remains difficult. To tackle the first challenge, we employed co-training and label propagation, two semi-supervised learning methods, to learn from a small amount of labeled data and a large pool of unlabeled data, thereby improving the accuracy of sea ice classifiers despite limited labeled data. To address the second challenge, we leveraged co-training’s built-in ability to integrate multiple data sources during the training process for the small labeled data. Additionally, we further enhanced data integration by using an ensemble of these co-trained models after training. Our approach demonstrates significant improvements over traditional supervised methods, showcasing the potential of semi-supervised learning methods in addressing two major challenges in developing automated sea ice classification solutions. Our study shows that semi-supervised learning improved F1 scores by 17% for SAR data and 33% for AMSR2 with limited labels, compared to supervised methods, while ensembling further boosted accuracy by 33%.

Sea ice, crucial to the Arctic and Earth's climate, requires consistent monitoring and high-resolution mapping. Manual sea ice mapping, however, is time-consuming and subjective, prompting the need for automated deep learning-based classification approaches. However, training these algorithms is challenging because expert-generated ice charts, commonly used as training data, do not map single ice types but instead map polygons with multiple ice types. Moreover, the distribution of various ice types in these charts is frequently imbalanced, resulting in a performance bias toward the dominant class. In this article, we present a novel GeoAI approach to training sea ice classification by formalizing it as a partial label learning task with explicit confidence scores to address multiple labels and class imbalance. We treat the polygon-level labels as candidate partial labels, assign the corresponding ice concentrations as confidence scores to each candidate label, and integrate them with focal loss to train a convolutional neural network. Our proposed approach leads to enhanced performance for sea ice classification in Sentinel-1 dual-polarized SAR images, improving classification accuracy (from 87% to 92%) and weighted average F-1 score (from 90% to 93%) compared to the conventional training approach of using one-hot encoded labels and categorical cross-entropy loss. It also improves the F-1 score in four out of the six sea ice classes.

AbstractAs Arctic sea ice and its overlying snow cover thin, more light penetrates into the ice and upper ocean, shifting the phenology of algal growth within the bottom of sea ice, with cascading impacts on higher trophic levels of the Arctic marine ecosystem. While field data or autonomous observatories provide direct measurements of the coupled sea ice‐algal system, they are limited in space and time. Satellite observations of key sea ice variables that control the amount of light penetrating through sea ice offer the possibility to map the under‐ice light field across the entire Arctic basin. This study provides the first satellite‐based estimates of potential sea ice‐associated algal bloom onset dates since the launch of CryoSat‐2 and explores how a changing snowpack may have shifted bloom onset timings over the last four decades.

Due to the growing volume of remote sensing data and the low latency required for safe marine navigation, machine learning (ML) algorithms are being developed to accelerate sea ice chart generation, currently a manual interpretation task. However, the low signal-to-noise ratio of the freely available Sentinel-1 Synthetic Aperture Radar (SAR) imagery, the ambiguity of backscatter signals for ice types, and the scarcity of open-source high-resolution labelled data makes automating sea ice mapping challenging. We use Extreme Earth version 2, a high-resolution benchmark dataset generated for ML training and evaluation, to investigate the effectiveness of ML for automated sea ice mapping. Our customized pipeline combines ResNets and Atrous Spatial Pyramid Pooling for SAR image segmentation. We investigate the performance of our model for: i) binary classification of sea ice and open water in a segmentation framework; and ii) a multiclass segmentation of five sea ice types. For binary ice-water classification, models trained with our largest training set have weighted F1 scores all greater than 0.95 for January and July test scenes. Specifically, the median weighted F1 score was 0.98, indicating high performance for both months. By comparison, a competitive baseline U-Net has a weighted average F1 score of ranging from 0.92 to 0.94 (median 0.93) for July, and 0.97 to 0.98 (median 0.97) for January. Multiclass ice type classification is more challenging, and even though our models achieve 2% improvement in weighted F1 average compared to the baseline U-Net, test weighted F1 is generally between 0.6 and 0.80. Our approach can efficiently segment full SAR scenes in one run, is faster than the baseline U-Net, retains spatial resolution and dimension, and is more robust against noise compared to approaches that rely on patch classification.

The formation of ice in clouds is an important process in mixed-phase clouds, and the radiative properties and dynamical developments of clouds strongly depend on their partitioning between the liquid and ice phases. In this study, we investigated the sensitivities of the cloud phase to the ice-nucleating particle (INP) concentration and thermodynamics. Moreover, passive satellite retrieval algorithms and cloud products were evaluated to identify whether they could detect cloud microphysical and thermodynamical perturbations. Experiments were conducted using the ICOsahedral Nonhydrostatic (ICON) model at the convection-permitting resolution of about 1.2 km on a domain covering significant parts of central Europe, and they were compared to two different retrieval products based on Spinning Enhanced Visible and InfraRed Imager (SEVIRI) measurements. We selected a day with multiple isolated deep convective clouds, reaching a homogeneous freezing temperature at the cloud top. The simulated cloud liquid pixel fractions were found to decrease with increasing INP concentration, both within clouds and at the cloud top. The decrease in the cloud liquid pixel fraction was not monotonic and was stronger in high-INP cases. Cloud-top glaciation temperatures shifted toward warmer temperatures with an increasing INP concentration by as much as 8 ∘C. Moreover, the impact of the INP concentration on cloud-phase partitioning was more pronounced at the cloud top than within the cloud. Furthermore, initial and lateral boundary temperature fields were perturbed with increasing and decreasing temperature increments from 0 to ±3 and ±5 K between 3 and 12 km, respectively. Perturbing the initial thermodynamic state was also found to systematically affect the cloud-phase distribution. However, the simulated cloud-top liquid pixel fraction, diagnosed using radiative transfer simulations as input to a satellite forward operator and two different satellite remote-sensing retrieval algorithms, deviated from one of the satellite products regardless of perturbations in the INP concentration or the initial thermodynamic state for warmer subzero temperatures while agreeing with the other retrieval scheme much better, in particular for the high-INP and high-CAPE (convective available potential energy) scenarios. Perturbing the initial thermodynamic state, which artificially increases the instability of the mid- and upper-troposphere, brought the simulated cloud-top liquid pixel fraction closer to the satellite observations, especially in the warmer mixed-phase temperature range.

The formation of orographic precipitation in mixed-phase clouds depends on a complex interplay of processes. This article investigates the microphysical response of orographic precipitation to perturbations of temperature and cloud condensation nuclei (CCN) concentration. A case study for the 2015 Cumbria flood in northern England is performed with sensitivities using a realization of the “piggybacking” method implemented into a limited-area setup of the Icosahedral Nonhydrostatic (ICON) model. A 6 % K−1 enhancement of precipitation results for the highest altitudes, caused by a “mixed-phase seeder–feeder mechanism”, i.e. the interplay of melting and accretion. Total 24 h precipitation is found to increase by only 2 % K−1, significantly less than the 7 % K−1 increase in atmospheric water vapour. A rain budget analysis reveals that the negative temperature sensitivity of the condensation ratio and the increase in rain evaporation dampen the precipitation enhancement. Decreasing the CCN concentration speeds up the microphysical processing, which leads to an increase in total precipitation. At low CCN concentration the precipitation sensitivity to temperature is systematically smaller. It is shown that the CCN and temperature sensitivities are to a large extent independent (with a ±3 % relative error) and additive.

The impact of cloud condensation nuclei (CCN) concentration on microphysical processes within thunderstorms and the resulting surface precipitation is not fully understood yet. In this work, an analysis of the microphysical pathways occurring in these clouds is proposed to systematically investigate and understand these sensitivities. Thunderstorms were simulated using convection-permitting (1 km horizontal grid spacing) idealized simulations with the ICON model, which included a 2-moment microphysics parameterization. Cloud condensation nuclei concentrations were increased from 100 to 3,200 CCN/cm(3), in five different wind shear environments ranging from 18 to 50 m/s. Large and systematic decreases of surface precipitation (up to 35%) and hail (up to 90%) were found as CCN was increased. Wind shear changes the details, but not the sign, of the sensitivity to CCN. The microphysical process rates were tracked throughout each simulation, closing the mass budget for each hydrometeor class, and collected together into "microphysical pathways, " which quantify the different growth processes leading to surface precipitation. Almost all surface precipitation occurred through the mixed-phase pathway, where graupel and hail grow by riming and later melt as they fall to the surface. The mixed-phase pathway is sensitive to CCN concentration changes as a result of changes to the riming rate, which were systematically evaluated. Supercooled water content was almost insensitive to increasing CCN concentration, but decreased cloud drop size led to a large reduction in the riming efficiency (from 0.79 to 0.24) between supercooled cloud drops and graupel or hail, resulting in less surface precipitation.

Arctic rain on snow (ROS) deposits liquid water onto existing snowpacks. Upon refreezing, this can form icy crusts at the surface or within the snowpack. By altering radar backscatter and microwave emissivity, ROS over sea ice can influence the accuracy of sea ice variables retrieved from satellite radar altimetry, scatterometers, and passive microwave radiometers. During the Arctic Ocean MOSAiC (Multidisciplinary drifting Observatory for the Study of Arctic Climate) expedition, there was an unprecedented opportunity to observe a ROS event using in situ active and passive microwave instruments similar to those deployed on satellite platforms. During liquid water accumulation in the snowpack from rain and increased melt, there was a 4-fold decrease in radar energy returned at Ku- and Ka-bands. After the snowpack refroze and ice layers formed, this decrease was followed by a 6-fold increase in returned energy. Besides altering the radar backscatter, analysis of the returned waveforms shows the waveform shape changed in response to rain and refreezing. Microwave emissivity at 19 and 89 GHz increased with increasing liquid water content and decreased as the snowpack refroze, yet subsequent ice layers altered the polarization difference. Corresponding analysis of the CryoSat-2 waveform shape and backscatter as well as AMSR2 brightness temperatures further shows that the rain and refreeze were significant enough to impact satellite returns. Our analysis provides the first detailed in situ analysis of the impacts of ROS and subsequent refreezing on both active and passive microwave observations, providing important baseline knowledge for detecting ROS over sea ice and assessing their impacts on satellite-derived sea ice variables.

Given growing interest in extreme high-latitude weather events, we use records from nine meteorological stations and atmospheric reanalysis data to examine extreme daily precipitation events (leading, 99th and 95th percentile) over Arctic Canada. Leading events span 90 mm at Cape Dyer, along the southeast coast of Baffin Island, to 26 mm at Sachs Harbour, on the southwest coast of Banks Island. The 95th percentiles range from 20 to 30% of leading event sizes. Extreme events are most common on or near the month of climatological peak precipitation. Contrasting with Eurasian continental sites having a July precipitation peak corresponding to the seasonal peak in precipitable water, seasonal cycles in precipitation and the frequency of extremes over Arctic Canada are more varied, reflecting marine influences. At Cape Dyer and Clyde River, mean precipitation and the frequency of extremes peak in October when the atmosphere is quickly cooling, promoting strong evaporation from Baffin Bay. At all stations, leading events involved snowfall and strong winds and were associated with cyclone passages (mostly of relatively strong storms). They also involved strong vapour fluxes, sometimes associated with atmospheric rivers or their remnants. The most unusual sequence of events identified here occurred at Clyde River, where the three largest recorded precipitation events occurred in April of 1977. Obtaining first-hand accounts of this series of events has proven elusive. Identified links between extreme events and atmospheric rivers demonstrates the need to better understand how the characteristics of such features will change in the future.

The mechanisms underlying decadal variability in Arctic sea ice remain actively debated. Here, we show that variability in boreal biomass burning (BB) emissions strongly influences simulated Arctic sea ice on multidecadal time scales. In particular, we find that a strong acceleration in sea ice decline in the early 21st century in the Community Earth System Model version 2 (CESM2) is related to increased variability in prescribed BB emissions in the sixth phase of the Coupled Model Intercomparison Project (CMIP6) through summertime aerosol-cloud interactions. Furthermore, we find that more than half of the reported improvement in sea ice sensitivity to CO 2 emissions and global warming from CMIP5 to CMIP6 can be attributed to the increased BB variability, at least in the CESM. These results highlight a new kind of uncertainty that needs to be considered when incorporating new observational data into model forcing while also raising questions about the role of BB emissions on the observed Arctic sea ice loss.

Abstract. Orographic precipitation is a key driver of flooding in mountainous areas. This article investigates the microphysical response of orographic rainfall to perturbations of temperature and cloud condensation nuclei (CCN) concentration. The study is motivated by the increased water vapour capacity of the atmosphere in a warming climate and the increasing frequency of extreme rainfall events. A case study for the Cumbria flood in December 2015 is performed with sensitivities using a realization of the `piggybacking' method implemented into a limited-area setup of the ICON model. A 6 % K-1 enhancement of rainfall results for the highest altitudes, caused by a `mixed-phase seeder-feeder mechanism', i.e. the interplay of melting and accretion. Total 24 h rainfall is found to increase by only 2 % K-1, significantly less than the 7 % K-1 increase in atmospheric water vapour. A rain budget analysis reveals that the negative temperature sensitivity of the condensation ratio and the increase of rain evaporation dampen the rainfall enhancement. Decreasing the CCN concentration speeds up the microphysical processing, which leads to an increase in total rainfall. At low CCN concentration the rainfall sensitivity to temperature is systematically smaller. It is shown that the CCN and temperature sensitivities are to a large extent independent (with a ±3 % relative error) and additive.

Abstract. Arctic rain-on-snow (ROS) deposits liquid water onto existing snowpacks. Upon refreezing, this can form icy crusts at the surface or within the snowpack. By altering radar backscatter and microwave emissivity, ROS over sea ice can influence the accuracy of sea ice variables retrieved from satellite radar altimetry, scatterometers, and passive microwave radiometers. During the Arctic Ocean MOSAiC Expedition, there was an unprecedented opportunity to observe a ROS event using in situ active and passive microwave instruments similar to those deployed on satellite platforms. During liquid water accumulation in the snowpack, there was a four-fold decrease in radar energy returned at Ku- and Ka-bands. After the snowpack refroze and ice layers formed, this decrease was followed by a six-fold increase in returned energy. Besides altering the radar backscatter, analysis of the returned waveforms shows the waveform shape changed in response to rain and refreezing. Microwave emissivity at 19 and 89 GHz increased with increasing liquid water content and decreased as the snowpack refroze, yet subsequent ice layers altered the polarization difference. Corresponding analysis of CryoSat-2 waveform shape and backscatter as well as AMSR2 brightness temperatures further shows the rain/refreeze was significant enough to impact satellite returns. Our analysis provides the first detailed in situ analysis of the impacts of ROS and subsequent refreezing on both active and passive microwave observations, providing important baseline knowledge for detecting ROS over sea ice and assessing their impacts on satellite-derived sea ice variables.

This study presents results from a model intercomparison project, focusing on the range of responses in deep convective cloud updrafts to varying cloud condensation nuclei (CCN) concentrations among seven state-of-the-art cloud-resolving models. Simulations of scattered convective clouds near Houston, Texas, are conducted, after being initialized with both relatively low and high CCN concentrations. Deep convective updrafts are identified, and trends in the updraft intensity and frequency are assessed. The factors contributing to the vertical velocity tendencies are examined to identify the physical processes associated with the CCN-induced updraft changes. The models show several consistent trends. In general, the changes between the High-CCN and Low-CCN simulations in updraft magnitudes throughout the depth of the troposphere are within 15% for all of the models. All models produce stronger (~+5%–15%) mean updrafts from ~4–7 km above ground level (AGL) in the High-CCN simulations, followed by a waning response up to ~8 km AGL in most of the models. Thermal buoyancy was more sensitive than condensate loading to varying CCN concentrations in most of the models and more impactful in the mean updraft responses. However, there are also differences between the models. The change in the amount of deep convective updrafts varies significantly. Furthermore, approximately half the models demonstrate neutral-to-weaker (~−5% to 0%) updrafts above ~8 km AGL, while the other models show stronger (~+10%) updrafts in the High-CCN simulations. The combination of the CCN-induced impacts on the buoyancy and vertical perturbation pressure gradient terms better explains these middle- and upper-tropospheric updraft trends than the buoyancy terms alone.

The use of the Global Fire Emissions Database (GFED) from 1997-2014 to create the CMIP6 historical biomass burning (BB) forcing allows for a more accurate representation of BB emissions in climate models, but also results in an unrealistic increase in their inter-annual variability compared to pre- and post-GFED years, especially in the Northern Hemisphere mid-latitudes. We find that this new BB forcing affects the simulated Arctic sea ice loss in several CMIP6 models, bringing them into better agreement with the observed sea ice decline by leading to enhanced sea ice loss in the early 21st century. This suggests that BB emissions may have played a role in the acceleration of the observed early 21st century Arctic sea ice loss. Using the Community Earth System Model version 2 (CESM2), we conduct sensitivity experiments in which we use BB emissions with a fixed annual cycle over the GFED period, to remove the inter-annual variability between 40-70°N. These experiments show that the strong acceleration in sea ice decline since the late 1990s simulated by the CESM2 is caused by enhanced Arctic warming driven by the increased variability in BB emissions over the GFED period. We also find that about half of the increase in sea ice sensitivity to CO2 and global mean surface temperature in the CESM2 compared to its CMIP5 counterpart, the CESM1, can be attributed to the change in BB emissions from CMIP5 to CMIP6, which suggests that the previously found improvement in sea ice sensitivity in CMIP6 models may in part be due to this new BB forcing and not only to changes in model physics. Overall, the results from this analysis highlight the influence of mid-latitude BB emissions on Arctic sea ice and provide new insights into the potential of a forced contribution to the observed accelerated early 21st century Arctic sea ice loss. Furthermore, this work highlights the importance of avoiding temporal discontinuities in prescribed aerosol forcing datasets as well as the need to better understand inter-model contrasts within the CMIP6 archive related to sensitivity to BB emissions.