Climate change is driving more frequent and severe wildfires in northwestern North American boreal forests, initiating shifts from conifer to broadleaf deciduous forest dominance. The resulting forests sequester more carbon and are more resistant to burning. However, when deciduous forests do burn, patterns and drivers of carbon losses are important for predicting long-term carbon storage in boreal forest landscapes. Here we use a combination of field and statistical modelling approaches to quantify carbon combustion losses in burned deciduous boreal forests. On average, deciduous forests lose less than half as much carbon to wildfire combustion as conifer forests per unit burned area. Although deciduous stands are more sensitive to top–down fire weather drivers than conifer stands, carbon loss is always lower than the minimum for conifer stands. This, along with the fire-suppressive effects of deciduous stands, could slow the positive feedback between wildfire and climate in fire-prone boreal landscapes. More frequent fires in the North American boreal are causing shifts from conifer to deciduous forests. This study finds that when deciduous forests burn, their carbon losses are driven by weather, but are lower than in conifer forests, potentially dampening climate–fire feedbacks.

This code is deemed central for the anaysis in the paper submitted as Brief report to PNAS: Canadian wildfires are losing their climate-cooling influence from post-fire snow albedo.

Methane is a potent greenhouse gas, which traps significantly more heat than carbon dioxide over short timescales, but remains underrepresented in models used for projecting future climate scenarios, particularly in permafrost landscapes. Understanding methane dynamics in northern wetlands is critical for improving projections of high-latitude carbon feedbacks to the global climate system. To address this, we integrated ecological processes driving methane production, oxidation, and transport pathways into the DVM-DOS-TEM terrestrial biosphere model. Using long-term site observations from a thermokarst bog in boreal Alaska, we calibrated and benchmarked the model against observations of methane and carbon dioxide fluxes, carbon and nitrogen stocks, and soil temperature and moisture. Parameter sensitivity analyses identified opposing correlations for methanogenesis and heterotrophic respiration, revealing a dependence on carbon availability. We also found sensitivity to the distribution of carbon relative to the water table position. Investigation of the dominance of different methane transport pathways demonstrated the need for observations of methane flux partitioning, and the utility in comparing simulated to observed seasonality. The cumulative methane efflux projected to the year 2100 had a range of 211-3470 g C m-2 between low and high warming scenarios. Methane emissions dominated by diffusion showed the greatest variability across projections. Increased soil temperature and carbon availability from permafrost thaw resulted in increasing methane emissions between 2030 and 2060. But methane emission was then limited by a deepening of the water table. Nevertheless, we estimated a positive radiative forcing (i.e., warming) from these mid-century methane emissions that persisted until 2100. An ebullition-dominant parameterization led to lower variability but a net negative radiative forcing (i.e., cooling) on average. Our study highlights the importance of representing methane emission pathways and the uncertainty associated with partitioning them on predicting the carbon budget and radiative forcing of wetland ecosystems in high latitudes.

Several of the largest amplifying feedbacks in the climate system—warming-induced emissions (WIE) of greenhouse gases from natural sources—remain absent from most climate models. Their absence risks systematic underestimation of projected warming. To assess the potential magnitude of these feedbacks, we derived relationships between global temperature and WIE rates from process-based model estimates across three Shared Socioeconomic Pathways, SSP1-2.6, SSP2-4.5, and SSP4-6.0. The warming-induced methane emission rate from permafrost, wetlands, freshwaters, and wildfire combined increases with temperature at 97 ± 6 Tg CH4 yr-1 ℃-1 across all three scenarios, while the warming-induced carbon dioxide emission rate from permafrost and wildfire combined increases at 7 ± 1 Pg CO2 yr-1 ℃-1 in SSP2-4.5 and SSP4-6.0. Using the MAGICC climate model, we projected that WIE could add 0.2–0.4 ℃ of warming by 2100 across scenarios, split roughly equally between carbon dioxide and methane, amplifying anthropogenic warming by 20–30%. Combined emission sensitivity and climate response uncertainties are ±0.2 ℃ on the added warming and ±10–30% on the warming amplification. Until Earth system models comprehensively incorporate WIE, climate projections are likely to underestimate future warming and overestimate remaining carbon budgets.

The 2023 Canadian fire season was record-breaking in terms of burned area and carbon emissions. Here, we present estimates of the regional climate-cooling effect from postfire surface albedo changes, which have historically partially offset the warming influence of fire emissions by wildfires. We estimate that the 2023 fires generated a time-integrated climate cooling of -3.41 W m-2 of burned area (95% CI: -4.39 to -2.43) over a 70-y period. We show that the climate-cooling impact has weakened on average by 29% since the 1960s due to changes in snow cover and duration. Collectively, this result implies that modern-day boreal fires are on average twice as likely to result in a net climate-warming influence.

Measurements of surface-atmosphere carbon dioxide (CO2) and methane (CH4) fluxes have been relatively sparse across the Arctic tundra and boreal biomes, causing significant uncertainties in carbon budget estimates from the region. While the availability of Arctic-boreal carbon flux data has increased substantially over the past decade, the data have remained spread across different repositories, scientific articles, and unpublished sources, making it difficult to leverage. Here we present a new dataset of monthly Arctic-boreal carbon fluxes (ABCFlux v2) across terrestrial (wetlands and uplands) and freshwater (lakes and rivers) ecosystems compiled from previous syntheses including the Arctic-boreal CO2 flux database (ABCFlux v1), the Boreal-Arctic Wetland and Lake Methane Dataset (BAWLD-CH4), and the Global River Methane Database (GRiMeDB). In addition, we consider data from general-purpose (e.g., Zenodo) and flux network repositories, literature, and site principal investigators. The dataset includes surface-atmosphere CO2 fluxes of gross primary production (GPP), ecosystem respiration (Reco), and net ecosystem exchange (NEE), alongside CH4 fluxes. For aquatic ecosystems, we split CH4 fluxes into diffusive and ebullitive flux pathways, and included potential emissions from transient storage in the water column (“storage fluxes”), alongside CO2 and CH4 concentrations dissolved in the surface water. Fluxes are measured through a variety of methods including chamber and eddy covariance techniques alongside bubble traps, ice-surveys, and concentration-based turbulence-driven modelling in aquatic ecosystems. The monthly flux data are reported together with supporting methodological and environmental metadata. The resulting ABCFlux v2 has 23 847 flux site-months, 8182 concentration site-months, and 199 seasonal observations from 1024 sites, and includes 56 139 reported fluxes (i.e. sum of GPP, Reco, NEE, and CH4 fluxes) from the years 1984 to 2024. The majority of monthly observations occurred after 1999. Wetlands had the highest number of site-month observations (8758), followed by boreal forest (6981), lotic ecosystems (6275), lentic ecosystems (3799) and upland tundra (3308). Measurements of CO2 dominated the dataset across most ecosystem types (25 222) except for lentic ecosystems, where CH4 flux site-months (3098) were more frequent than CO2 flux site-months (2915). Overall, ABCFlux v2 includes 160 % more site-months for terrestrial CO2 flux data compared to ABCFlux v1. Integrating and updating BAWLD-CH4 flux data from growing season averages to monthly fluxes resulted in 5671 site-months of chamber CH4 data compared to 762 site-years. This collaborative initiative, involving contributions from over 260 researchers, provides a comprehensive overview of the current state of the Arctic-boreal carbon flux network and its data, and serves as an important step in reducing uncertainties in Arctic-boreal carbon budgets and in enhancing our understanding of climate feedbacks. The data can be accessed at ORNL DAAC at https://doi.org/10.3334/ORNLDAAC/2448 (Virkkala et al., 2026).

The boreal forest biome is warming rapidly, impacting disturbance regimes and global climate. Boreal forest fires have intensified, initiating both climate warming (positive) and climate cooling (negative) impacts across spatial and temporal scales. Here we estimate climate impacts from boreal fires in Alaska and western Canada between 2001 and 2019 using integrated net radiative forcing metrics combining greenhouse gas and aerosol emissions from combustion, vegetation recovery, greenhouse gas emissions from fire-induced permafrost thaw and changes in surface albedo over a 70-year period. We find that fires across Alaska contributed, on average, to net climate warming (0.35 ± 4.66 W m-2 of burned area; one standard deviation), while fires across Canada contributed to net cooling (-2.88 ± 4.17 W m-2 of burned area; one standard deviation). Climate-warming fires occur preferentially in dry, high-elevation, steep permafrost landscapes with high pre-fire black spruce coverage and combust more carbon per unit area. Climate-cooling fires are driven by longer spring snow exposure and occur more frequently in continental regions near the treeline. This fine-scale characterization of component and net radiative forcing advances our understanding of the biogeophysical impacts of fires on high-latitude climate and highlights the need to prioritize fire management in carbon-rich permafrost regions to curb long-term warming.

Wildfires in the Arctic-boreal zone have increased in frequency over recent decades, carrying substantial ecological, social, and economic consequences. Remote sensing is crucial for mapping burned areas, monitoring wildfire dynamics, and evaluating their impacts. However, existing high-latitude burned area products suffer from significant discrepancies, particularly in Siberia, and their coarse spatial resolutions limit accuracy and utility. To address these gaps, we developed a convolutional neural network model to map burned areas at a 30 m resolution across the Arctic- boreal zone using Landsat and Sentinel-2 imagery. Using vegetation indices including the normalized burn ratio, normalized difference vegetation index, and normalized difference infrared index our model achieved strong performance, with an Intersection Over Union (IOU) of 0.77 and an F1 score of 0.85 on unseen test data. Performance was higher in North America (IOU = 0.84) than in Eurasia (IOU = 0.72), reflecting regional differences in fire regimes and data quality. Predictions for six representative years showed our model's burned area closely matched the median values of Landsat, MODIS, and VIIRS-based products, although alignment varied annually and spatially. Visual assessments indicated our approach was generally more accurate, notably in detecting unburned vegetation islands within fire perimeters missed by other products. This research has numerous potential applications, such as analysing feedback between vegetation and burn patterns, characterizing spatial dynamics of unburned islands, and improving carbon emission estimates through detailed burn severity assessments.

Accurate delineations of Earth’s surface features are crucial for environmental science, but for dynamic landscapes like thawing permafrost, these datasets rapidly become outdated and inaccurate. Manual refinement of these outdated labels is prohibitively slow, while existing foundation segmentation models often fail to accurately segment complex natural features in satellite images. To address this, we aim to use inaccurate labels as geospatial priors to prompt a vision foundation model for zero-shot semantic segmentation of Earth observation objects with current satellite imagery. We developed an interactive segmentation framework that synergistically combines the Segment Anything Model (SAM) with expert oversight. Our system features two components: (1) a scalable automated pipeline that uses historical delineations as a strong geospatial prior for prompting SAM, and (2) an interactive tool that enables experts to refine outputs and create new, high-fidelity labels rapidly. Using the challenging case study of retrogressive thaw slumps, we demonstrate that this human–AI partnership achieves robust segmentation performance, with a best Intersection over Union of 0.80 ± 0.10 for routine updates. This approach matches the quality of manual expert digitisation but at twice the efficiency. This model- and task-agnostic framework offers a practical, transferable solution for generating the dense, time-series data required to monitor landscape changes, transforming a laborious mapping task into an AI-assisted, expert-overseen workflow.

Rapid Arctic warming is thawing carbon-rich permafrost, releasing greenhouse gases that accelerate climate change. Despite the importance of this feedback, permafrost-enabled global-scale models simulate only gradual, top-down thickening of the seasonally-thawed soil. This ignores abrupt permafrost thaw and intensifying fire regimes that combust soil carbon and further accelerate thaw. Here, we expand a compact Earth system model (OSCAR v3.0) enabling initial estimates of the impacts of abrupt thaw and wildfire, together with gradual thaw, on remaining carbon budgets consistent with the temperature goals of the Paris Agreement. Our model suggests that including permafrost thaw and fire-related carbon emissions reduces the remaining allowable carbon budgets from 2025 onward by 25

Warming-induced emissions (WIE) of greenhouse gases from natural sources are a large amplifying feedback that most climate models do not yet represent. This absence risks systematic underestimation of projected future warming and overestimation of remaining carbon budgets. To assess the potential magnitude of these feedbacks, we derived relationships between global temperature and WIE rates from process-based model estimates across three Shared Socioeconomic Pathways, SSP1-2.6, SSP2-4.5, and SSP4-6.0. The warming-induced methane emission rate from permafrost, wetlands, freshwaters, and wildfire combined increases with temperature at an estimated 97 ± 6 Tg CH _4 yr ^−1 °C ^−1 across all three scenarios, while the warming-induced carbon dioxide emission rate from permafrost and wildfire combined increases at 7 ± 1 Pg CO _2 yr ^−1 °C ^−1 in SSP2-4.5 and SSP4-6.0. Using the MAGICC climate model, we projected that WIE could add 0.2 °C–0.4 °C of warming by 2100 across scenarios, split roughly equally between carbon dioxide and methane, amplifying post-2020 anthropogenic warming by 20%–30%. Combined emission sensitivity and climate response 1 σ uncertainties are ±0.2 °C on the added warming and ±10%–30% on the warming amplification.

Vegetation growth in the high altitudinal region such as the Qinghai‒Tibet plateau (QTP) is generally considered to be limited by air temperature, and warming has been reported to drive widespread vegetation greening. However, warming may also exacerbate soil water deficit and accelerate thermokarst development in the permafrost region. The extent to which these processes influence vegetation growth in the QTP remains unclear. Here, we assessed recent vegetation changes in the Three-Rivers Headwaters Region (TRHR), a representative permafrost region in the central and eastern QTP with ecological significance, through reconstructing a fine-resolution (30-m) Landsat vegetation greenness time series spanning 2000–2023. Our results indicate that recent warming and thermokarst development has caused a complex vegetation browning and greening pattern across the TRHR. Substantial (11.1%–15.2%) vegetation browning was observed at dense vegetation areas at lower-elevation non-permafrost zone, mainly attributed to warming induced soil water deficit. Conversely, dominant vegetation greening (44.6%) and limited (4.2%) vegetation browning was observed at higher-elevation permafrost zone with more sparse vegetation covers. Moreover, in the permafrost areas, thermokarst landscapes are becoming hotspots of vegetation greening and browning. Thermokarst lake areas show strong vegetation greening trends, with largest positive vegetation greenness trends (0.32%–0.34% per year) in recently drained lake areas, while retrogressive thaw slump areas show abrupt vegetation losses. With continued strong warming, increasing soil water limitation and intensifying thermokarst activity will likely create a more complex vegetation greening and browning pattern in the QTP region, implying vulnerable permafrost and ecosystem stability under global warming with potentially large climate feedbacks.

Bioclimatic feedbacks, driven by anthropogenic warming, produce indirectly human-caused emissions that contribute to a significant discrepancy between models, quantification frameworks, and atmospheric change. This discrepancy threatens climate ambitions, but policy is ill equipped to quantify the threat. Under two scenarios (SSP 1-1.9, roughly aligned with Paris Agreement goals and SSP 2-4.5, roughly current trajectory), thawing permafrost, increasing circumboreal fires, and warming tropical wetlands increase CO2 emissions by '15 to '300 Tg C year-1 and CH4 by '17 to '50 Tg C year-1 above 2020 levels by 2050, lessening time to the 1.5 degrees C and 2.0 degrees C Paris thresholds by '21%-25%. Policy frameworks and tools for quantifying/reporting indirect emissions from managed and unmanaged lands should be developed. The Earth system modeling community could inform this effort and would benefit from additional data. Ultimately, increased mitigation ambition to compensate for indirect emissions will likely proceed only if processes are measured and reported.

Arctic and boreal regions (ABRs) are experiencing rapid warming and increasingly severe wildfires, threatening their roles as global carbon sinks. High quality time series maps of aboveground biomass (AGB) are key for characterizing and attributing spatiotemporal dynamics of carbon stocks in these regions. However, existing maps at regional to global scales often lack the spatial resolution or temporal coverage needed to capture the heterogeneous and dynamic nature of Arctic-boreal AGB change. To address these limitations, we developed annual (1984-2022) 30-m resolution AGB density maps for Alaska and Canada (11.2 x 10(6) km(2)). The maps were produced by using extensive training datasets, including 45,002 unique ground plots and 100,000 km(2) of airborne lidar data, and time-series spectral features derived from the Continuous Change Detection and Classification (CCDC) algorithm fitted on Landsat Collection 2 Surface Reflectance. Using the eXtreme Gradient Boosting model, we generated annual wall-to-wall maps of AGB along with associated uncertainties. Our maps suggest a similar to 41 Pg stock of AGB at 2022, representing a 12% increase from 1984. Our maps achieve high accuracy and low bias on holdout testing data (R-2 = 0.72, Bias= -5.03%, RMSE% = 62.7%), representing an average of 16.0 percentage points increase in R-2 values and 22.7 percentage points decrease in relative bias compared to six existing AGB products. We show using repeat ground measurements that these maps effectively capture AGB loss and recovery due to fire and harvest, gradual AGB changes, and both live and dead tree AGB components in boreal regions. By integrating extensive calibration data with multi-decadal satellite observations and advanced machine learning techniques, these map products provide a robust tool for advancing the understanding of carbon dynamics under global change in Arctic-boreal North America.

Fire is a global phenomenon and a key Earth system process. Extreme fire events have increased in recent years, and fire frequency and intensity are projected to rise across most regions and biomes, posing substantial challenges for ecosystems, the carbon cycle, and society. The Fire Model Intercomparison Project (FireMIP), launched in 2014, has advanced global fire modeling in Dynamic Global Vegetation Models (DGVMs) and improved understanding of fire's local and direct drivers and its local impacts on vegetation and land carbon budgets through land offline simulations (i.e., uncoupled from the atmosphere). We now bring FireMIP into Coupled Model Intercomparison Project Phase 7 (CMIP7) to: (1) evaluate fire simulations in state-of-the-art fully coupled Earth system models (ESMs); (2) assess fire regime changes in the past, present, and future, and identify their primary natural and anthropogenic forcings and causal pathways within the Earth system, including the associated uncertainties; and (3) quantify the impacts of fires and fire changes on climate, ecosystems, and society across Earth system components, regions, and timescales, and elucidate the underlying mechanisms. FireMIP in CMIP7 will advance the fire and fire-related modeling in fully coupled ESMs, and provide a quantitative, comprehensive, and process-based understanding of fire's role in the Earth system by using models that incorporate critical climate feedbacks and CMIP7 multi-model, multi-initial-condition, and multi-scenario ensemble. This protocol paper presents the motivation, scientific questions, experimental design and rationale, model inputs and outputs, and recommended analysis framework for FireMIP in CMIP7, providing guidance to Earth system modeling teams conducting simulations and informing communities studying fire, climate change, and climate solutions.

Arctic permafrost is undergoing rapid changes due to climate warming in high latitudes. Retrogressive thaw slumps (RTS) are one of the most abrupt and impactful thermal-denudation events that change Arctic landscapes and accelerate carbon feedbacks. Their spatial distribution remains poorly characterised due to time-intensive conventional mapping methods. While numerous RTS studies have published standalone digitisation datasets, the lack of a centralised, unified database has limited their utilisation, affecting the scale of RTS studies and the generalisation ability of deep learning models. To address this, we established the Arctic Retrogressive Thaw Slumps (ARTS) dataset containing 23,529 RTS-present and 20,434 RTS-absent digitisations from 20 standalone datasets. We also proposed a Data Curation Framework as a working standard for RTS digitisations. This dataset is designed to be comprehensive, accessible, contributable, and adaptable for various RTS-related studies. This dataset and its accompanying curation framework establish a foundation for enhanced collaboration in RTS research, facilitating standardised data sharing and comprehensive analyses across the Arctic permafrost research community.

The soil moisture active passive (SMAP) satellite mission distributes a product of CO2 flux estimates (SPL4CMDL) derived from a terrestrial carbon flux model, in which SMAP brightness temperatures are assimilated to update soil moisture (SM) and constrain the carbon cyclemodeling. While the SPL4CMDL product has demonstrated promising performance across the continental USA and Australia, a detailed assessment over the arctic and subarctic zones (ASZ) is still missing. In this study, SPL4CMDL net ecosystem exchange (NEE), gross primary production (GPP), and ecosystem respiration (R-E) are evaluated against measurements from 37 eddy covariance towers deployed over the ASZ, spanning from 2015 to 2022. The assessment indicates that the NEE unbiased root-mean-square error falls within the targeted accuracy of 1.6 gC.m(-2).d(-1), as defined for the SPL4CMDL product. However, modeled GPP and R-E are overestimated at the beginning of the growing season over evergreen needleleaf forests and shrublands, while being underestimated over grasslands. Discrepancies are also found in the annual net CO2 budgets. SM appears to have a minimal influence on the GPP and R-E modeling, suggesting that ASZ vegetation is rarely subjected to hydric stress, which contradicts some recent studies. These results highlight the need for further carbon cycle process understanding and model refinements to improve the SPL4CMDL CO2 flux estimatesover the ASZ.

Wetlands are the largest natural source of methane (CH4) emissions globally. Northern wetlands (>45° N), accounting for 42 % of global wetland area, are increasingly vulnerable to carbon loss, especially as CH4 emissions may accelerate under intensified high-latitude warming. However, the magnitude and spatial patterns of high-latitude CH4 emissions remain relatively uncertain. Here, we present estimates of daily CH4 fluxes obtained using a new machine learning-based wetland CH4 upscaling framework (WetCH4) that combines the most complete database of eddy-covariance (EC) observations available to date with satellite remote-sensing-informed observations of environmental conditions at 10 km resolution. The most important predictor variables included near-surface soil temperatures (top 40 cm), vegetation spectral reflectance, and soil moisture. Our results, modeled from 138 site years across 26 sites, had relatively strong predictive skill, with a mean R2 of 0.51 and 0.70 and a mean absolute error (MAE) of 30 and 27 nmol m−2 s−1 for daily and monthly fluxes, respectively. Based on the model results, we estimated an annual average of 22.8±2.4 Tg CH4 yr−1 for the northern wetland region (2016–2022), and total budgets ranged from 15.7 to 51.6 Tg CH4 yr−1, depending on wetland map extents. Although 88 % of the estimated CH4 budget occurred during the May–October period, a considerable amount (2.6±0.3 Tg CH4) occurred during winter. Regionally, the Western Siberian wetlands accounted for a majority (51 %) of the interannual variation in domain CH4 emissions. Overall, our results provide valuable new high-spatiotemporal-resolution information on the wetland emissions in the high-latitude carbon cycle. However, many key uncertainties remain, including those driven by wetland extent maps and soil moisture products and the incomplete spatial and temporal representativeness in the existing CH4 flux database; e.g., only 23 % of the sites operate outside of summer months, and flux towers do not exist or are greatly limited in many wetland regions. These uncertainties will need to be addressed by the science community to remove the bottlenecks currently limiting progress in CH4 detection and monitoring. The dataset can be found at https://doi.org/10.5281/zenodo.10802153 (Ying et al., 2024).

Wildfire frequency, intensity, and rate of spread are increasing across the Western U.S, resulting in more severe ecosystem impacts. Significant tree mortality can occur years after fire events, but this has received little attention compared to the immediate tree loss during a fire. We overlapped forest cover loss data with burn severity maps in the U.S. Pacific Northwest and quantified the total and delayed forest canopy loss after fires. We found that wildfires resulted in total canopy loss fraction (CLF) of 84%, 53%, and 22% within 3 years in areas burned at high, moderate, and low severity, respectively. The delayed canopy loss accounted for approximately 1/3, 1/2, and 2/3 of the total canopy loss for high, moderate, and low severity burns. Delayed canopy loss was greater in moist and cool areas than in dry and warm areas, likely because tree species in wetter environments were less adapted to survive when fires did occur. Across all forests, delayed CLF doubled as temperature increased from the climatological mean to a hot anomaly and tripled as vapor pressure deficit increased from a wet anomaly to a dry anomaly. Fire impacts on forest ecosystems are likely to intensify under future climate scenarios as wildfires expand into areas that historically experienced infrequent fires. The impacts can also be exacerbated by more frequent compound extreme events, such as droughts, heatwaves, and fires. These findings highlight the urgent need for targeted forest management strategies, particularly in mesic forests, to mitigate future fire impacts.

Retrogressive Thaw Slumps (RTS) in Arctic regions are distinct permafrost landforms with significant environmental impacts. Mapping these RTS is crucial because their appearance serves as a clear indication of permafrost thaw. However, their small scale compared to other landform features, vague boundaries, and spatiotemporal variation pose significant challenges for accurate detection. In this article, we extend a state-of-the-art deep learning model to delineate RTS features across the Arctic in a multimodal setting. Two new strategies were introduced to optimize multimodal learning and enhance the model's predictive performance: 1) a feature-level, residual cross-modality attention fusion strategy, which effectively integrates feature maps from multiple modalities to capture complementary information and improve the model's ability to understand complex patterns and relationships within the data; 2) pretrained unimodal learning followed by multimodal fine-tuning to alleviate high computing demand while achieving strong model performance. Experimental results demonstrated that our approach outperformed existing models adopting data-level fusion, feature-level convolutional fusion, and various attention fusion strategies, providing valuable insights into the efficient utilization of multimodal data for RTS mapping. This research contributes to our understanding of permafrost landforms and their environmental implications.