Millions of songbirds migrate to the Arctic each spring to breed. In the fall, these birds and their young disperse globally to overwintering grounds where they provide ecosystem services as prey, seed dispersers, nutrient transporters, and charismatic wildlife. In recent decades, however, the “Arctic songbird nursery” has experienced dramatic shifts in abiotic and biotic conditions, driven primarily by unprecedented warming, heightened climatic variability, and vegetation change. In order to understand how these changes may affect the success of Arctic songbird species, a quantitative understanding of how the size and diversity of migratory songbird communities responds to spatiotemporal variation in breeding habitat characteristics is needed. Here, we take preliminary steps towards that goal by synthesizing bioacoustically-derived estimates of songbird community composition with multi-source remote sensing observations of habitat conditions. To do this, we used a deep neural network to estimate the size, species composition, and relative abundance of Arctic songbird communities from bioacoustic data collected during the 2024 breeding season at 13 sites spanning a large gradient of latitudinal, environmental, and vegetation conditions on the North Slope of Alaska. We then used remote sensing imagery from Landsat, the Land, Vegetation, and Ice Sensor, Sentinel-2, and PlanetScope, as well as modeled climatic data from SnowModel, to obtain spatially explicit information on the sites’ topography, climate, snow dynamics, and vegetation phenology, composition, and structure. Finally, the data were synthesized to train and validate a multivariate random forest model capable of predicting current songbird community diversity and species assemblages from habitat characteristics at the sites, thereby quantifying how landscape-level variation in abiotic and biotic conditions influences songbird community diversity. In the future, insights gained from this small-scale application will be used to develop a larger model capable of predicting songbird community diversity across the broader western North American Arctic and under future climate, vegetation, and resource availability scenarios. Ultimately, these quantitative predictions of how songbirds respond to shifts in ecosystem form and function over space and time will advance our understanding of Arctic songbird biodiversity and its resilience to environmental change.
  [poster]

Major advances in understanding carbon emissions from boreal forest fires have been made in recent years, yet substantial knowledge gaps remain for tundra fires. Tundra ecosystems are less prone to fires than boreal forests, but are experiencing increasing fire activity due to climate change. Tundra organic soils store large amounts of carbon and insulate permafrost. Additionally, studies assessing the application of remotely sensed fire severity metrics to tundra fires are limited. Here, we report the carbon combustion from two tussock tundra fires that burned together 1047.6 km² in Southwest Alaska in 2022. We quantified above- and belowground carbon stocks, combustion rates, and thaw depth in 45 field plots (36 burned and 9 unburned) one year after the fire. We determined soil burn depth using paired moss patches as reference points and applied allometric equations to calculate carbon stocks from shrubs and tussocks. Within an additional 41 plots, fire severity was estimated using the Geometrically structured Composite Burn Index (GeoCBI). We upscaled and assessed our field-based findings using Sentinel-2 imagery. Tussock tundra landscapes emitted an average of 1.6 ± 0.6 kg C/m². The majority (74%) of the carbon lost was from belowground stocks, with a burn depth of 6.9 ± 2.1 cm. 11% of the total pre-fire carbon stock was consumed. We found mid-summer thaw depth on average 19.6 cm deeper in burned than in unburned sites. Using the differenced normalized burn ratio (dNBR), we scaled the carbon combustion (R² = 0.42, p < 0.001) to the burned area covered by tussock tundra (48.7% of the total). The two fires together emitted around 0.8 Tg of carbon from tussock tundra ecosystems. GeoCBI field estimates were highly correlated with dNBR (R² = 0.82, p < 0.001) across vegetation types. This work provides important information on the impacts of fires on the carbon balance of tundra ecosystems. We also showed that the dNBR, a well-established remote sensing proxy for forest fires, is an effective metric for mapping fire severity over tundra landscapes.
  [poster] 

In recent years, large wildfires have spread in Arctic regions as a consequence of ongoing climate change. Arctic organic soils are comparatively shallow but may be ancient, thus thousands of years old carbon may be released in smoldering and deeply burning fires. In Greenland, a land known for its icy expanse, fires are extremely rare. However, in summer 2019, the second-largest wildfire ever recorded on the island occurred at the Kangerluarsuk Tulleq fjord in southwestern Greenland. This study aims to produce pioneering in-field data on this tundra fire, focusing on two key aspects: 1) carbon loss and 2) the age of the carbon released. Understanding whether the released carbon is modern or old is crucial due to different implications for the global carbon cycle and climate. To estimate carbon losses from the Kangerluarsuk Tulleq tundra fire, we established 14 sampling plots in burned areas and at unburned control sites. The selection of sampling plots was guided by a differenced Normalized Burn Ratio (dNBR) map generated using Sentinel-2 data and field reconnaissance. Within each plot, we assessed fire severity to estimate the above-ground carbon loss. For below-ground carbon loss estimation and burn depth analysis, organic soil samples were collected at burned plots and compared with unburned ones. To explore the vegetation succession and burned vegetation type, organic soil profiles (n=10) were extracted down to the mineral ground using a soil box corer and were studied by stereo-microscopy. Subsamples (n=~30) from burned and unburned soil horizons were selected for radiocarbon dating to determine the age of carbon released in the fire. Our results suggest that soil carbon loss was higher than previously reported at an Alaskan tundra fire site with a mean value of 6.185 ± 1.2 kg of C m-2. The mean burn depth was 9.3 ± 1.4 cm, and soil thaw depths during the 2024 summer were approximately 24 cm deeper in the 2019 burned area compared to unburned tundra. Radiocarbon analysis showed that the maximum age of the carbon released by the fire was record-breaking 928 ± 67 calBP. Vegetation succession measurements showed that post-fire surfaces were predominantly colonized by pioneering non-Sphagnum bryophytes, Aulacomnium androgynum and A. turgidum. The acquired results are first of a kind from a Greenland tundra fire and produce essential data for global climate modeling to assess the climate impacts of increasing Arctic wildfires.
  [poster] 

Over the past decades, the boreal forest region has been warming rapidly, initiating complex feedback mechanisms affecting its carbon budget. While warming allows for a longer growing season, it also causes the thaw of carbon- and ice-rich permafrost soils. Simultaneously, boreal forests have been increasingly impacted by wildfire. In Interior Alaska, fire disturbance initially causes the mortality of conifers, removal of insulating organic soil layers, and reduction in surface albedo, followed by vegetation succession that includes a productive deciduous phase. Belowground, fires trigger a cycle of permafrost degradation and recovery. However, with warming and, or without the recovery of coniferous forests with a thick mossy understory, permafrost aggradation may be inhibited by the formation of talik (subsurface soil layers that remain unfrozen year-round). The presence of talik may greatly increase year-round losses of ancient carbon in the form of the greenhouse gas CO2 and further accelerate permafrost thaw. Advancing our ability to forecast the carbon budget of this rapidly changing region urgently requires more direct observations of how boreal forests and their soils are impacted by and recover from wildfires. Key uncertainties are the trajectories of soil temperature and moisture and the magnitude, sources (contributions from root vs. microbial respiration in the active layer and thawing permafrost), and timing (especially at the onset of the growing season and in early winter) of soil-atmosphere CO2 fluxes. To address this knowledge gap, we investigated the timing, magnitude, and sources of soil CO2 emissions at paired burned (2003) and unburned boreal forest sites near Hess Creek, Alaska. Alongside soil temperature and moisture, we measured soil-atmosphere CO2 fluxes (2016-2023) and their sources (2021-2022) year-round, using novel forced-diffusion-based soil respiration stations (n=3) and multi-depth passive 14CO2 samplers (n=2). We combined these data with bulk soil analysis and incubations and used an isotope mass balance approach to apportion CO2 sources. Preliminary findings show that fire resulted in a 60% reduction of organic layer thickness and a 100% increase in active layer thickness. There was no significant change in total annual soil-atmosphere CO2 emissions over time or between the burned and unburned site. However, fire significantly decreased the seasonal amplitude of soil-atmosphere CO2 emissions and led to higher emissions outside the growing season (Sep-Apr) and losses of older carbon pools. The new data produced by this effort will greatly advance our understanding of the impact of fire disturbance on permafrost soil properties and conditions that control year-around soil CO2 emissions from boreal ecosystems.

Background: Arctic and boreal regions (ABRs) are experiencing the most rapid warming, including increasing frequency and severity of disturbances such as wildfire. These changes may have converted some terrestrial ecosystems in ABRs into carbon sources, threatening their future role as a carbon sink despite observations of greening in these systems. High resolution time series biomass maps are key to revealing the temporal and spatial dynamics of aboveground biomass (AGB) and to improve our capability on projecting future carbon dynamics. Here we aim to produce high quality time series maps between 1984-2022 for North America (NA) ABRs, including Canada and Alaska, and to uncover the important drivers impacting AGB dynamics. Methods: To produce high quality time series AGB maps, we compiled 45,002 unique ground plots with 77,323 field measurements from various sources and more than 100,000 km2 of airborne LiDAR scanning data. We then trained a model to predict AGB per ecoregion using spectral-temporal features derived from Landsat imagery computed with the Continuous Change Detection and Classification algorithm. We projected this model to develop maps of AGB across Canada and Alaska for each year in 1984-2022 at 30 m spatial resolution. After deriving the time series AGB maps, we analyzed the important climatic drivers of AGB dynamics. Results: Our maps showed good overall accuracy on held-out testing data (R2= 0.76, Bias%= -5.6). Our map also showed good capability to capture temporal AGB dynamics and to capture both live and dead AGB. We find that the total AGB stock of the study domain increased from 56.79 to 64.26 Pg, an increase of 13.2%. AGB increased by 2%-22% across different ecoregions of study areas during 1985-2022. However, Northern Forests (NF) lost about -0.4% of AGB during 2015-2022. Based on our climatic driver analysis, higher average spring temperature is generally associated with faster rate of AGB gain, while higher average autumn temperature is generally associated with decreased AGB growth rate, which may be due to higher vapor pressure deficit and palmer drought severity index. Next steps: We will publish our annual AGB maps to ORNL DAAC in the year of 2025, and we will also conduct more thorough analysis of AGB drivers, including wildfire, harvest, and drought. We will also compare AGB budgets derived from our maps with estimates from governmental carbon accounting reports and earth system models.
 

Increased wildfire activity threatens the resilience of North American boreal forests by altering the composition, structure, and function of the ecosystem. Evidence suggests that boreal forest communities in burned areas are shifting from predominantly evergreen needleleaf forest to deciduous broadleaf and mixed forests, potentially altering ecosystem functions such as evapotranspiration. This space-for-time study investigates trends in boreal forest composition (land cover), structure (aboveground biomass), and function (evapotranspiration) during a three-decade period of regeneration following wildfire events. The ABoVE Annual Dominant Land Cover and Annual Aboveground Biomass products are used to measure vegetative recovery. Evapotranspiration (ET) is estimated using BorealET, a novel adaptation of the MOD16 algorithm that incorporates a dynamic land cover input (ABoVE Annual Dominant Land Cover) and estimates of LAI and fPAR from gap-filled optical imagery (HISTARFM). In validation, BorealET performed with lower RMSE and bias than MOD16 Collection 6.1 across 33 regional eddy covariance flux towers in the AmeriFlux network. Trends in ET were assessed across the study region with simulations to isolate the direct effects of climate change from the effects of land cover change. From 2001-2020, it was estimated that ET is generally increasing across the study region at a rate of 2.42 mm year-2, with three quarters of the trend (1.83 mm year-2) attributable to land cover change, suggesting that the inclusion of a dynamic land cover component is an important advancement in process-based ET models. Results of the space-for-time study confirm that burned areas in North American boreal forests are generally undergoing successional trajectories toward deciduous and mixed forests. Rates of ET, while negatively impacted immediately after wildfire events rebound and surpass pre-fire levels after about 20 years. Trends in all variables are sensitive to wildfire severity.
  [poster] 

Northern lakes provide habitat for year-round and migratory animals and regulate carbon, nutrient, and sediment movement. People also rely on northern lakes for drinking water, subsistence activities, and industrial operations. As such, monitoring long-term changes in lakes is crucial to understanding how humans and climate change impact these essential ecosystems. However, multidecadal trends in northern lake area are highly uncertain, with different studies reporting directionally opposite trends over the same region. Here, we examine the sources of differences between lake area estimates and short-term lake area trends derived from two Landsat- and one Sentinel-2-based surface water products across five northern study regions. We show that differences in the magnitude and direction of regional lake area trends are related to systematic between-product differences in surface water detection in dry vs. wet years. In particular, the differences between lake areas estimated with Landsat products and the Sentinel-2 product were larger in dry years compared with wet years. This between-product difference reflects how the products classify mixed shoreline and other ambiguous pixels, and was particularly large in regions with small, shallow lakes and emergent, floating, and/or aquatic vegetation. Our results suggest that multi-decadal lake area trends derived from optical satellite imagery are most uncertain in regions with a high density of these small, shallow lakes containing emergent, floating, and/or aquatic vegetation, and that clarifying the magnitude and direction of lake area trends will require new technologies and methods designed to differentiate between water, land, and inundated vegetation.
  [poster]

The rapid warming of the Arctic-Boreal zone drives profound changes in hydrological dynamics, with significant impacts on local communities, wildlife, and greenhouse gas (GHG) feedbacks. Small water bodies (<0.1 km2) are hydrologically sensitive, potent emitters of methane (CH4), yet their contribution to total GHG emissions is poorly quantified. To address this uncertainty, we used PlanetScope imagery (3.5 m) to map open water lakes and ponds as small as 100 m2 across the entire Yukon-Kuskokwim Delta, Alaska. Water body area distributions from the high-resolution maps informed a fully coupled thermal-biogeochemical lake model (LAKE) to estimate regional water body CH4 fluxes from 2002 – 2023. Multiple LAKE configurations were generated based on water body surface area. CH4 emissions for each surface area class were calibrated to match observed CH4 emissions from the literature. Model simulations demonstrate the significance of small water bodies in total CH4 emissions and revealed a 0.1 Tg/yr linear increase (r2 = 0.5; p < 0.01) in CH4 ebullition from all water bodies throughout the study period. In addition to open water lake and pond maps, we developed 3.5 m partial inundation maps to supplement future investigations. Although partial inundation was not accounted for in this modeling experiment, these areas are particularly important components of the CH4 budget. Future work will leverage both open water and partial inundation products to improve GHG emissions modeling. Additional work will incorporate these maps in assessments of community risk and waterfowl habitat.

Snow is a fundamental control on Arctic and subarctic processes, including surface and subsurface energy, water, and carbon balances. The distribution of Arctic snow is an important driver of subsurface thermal regimes, including permafrost temperatures, as deeper snow acts to insulate the subsurface like a blanket. Snow distribution also interacts with vegetation dynamics, leading to the entrainment of snow within shrub patches, increasing soil moisture and leading to feedbacks that can enhance shrub growth. As the climate shifts, an acceleration of the Arctic hydrologic cycle has led to increased snow and shrubification—expansion and enlargement of shrubs within Arctic ecosystems. However, Arctic processes related to subgrid-scale snow distribution are not well represented within Earth system models, leading to potential errors in both permafrost thaw and associated carbon dynamics. Here, we introduce a version of the land model component of the Energy Exascale Earth System Model (E3SM), ELM, with improved snow representation for Arctic ecosystems. Through intercomparisons with intensive in-situ observations, we evaluate ELM’s configurations using snow data collected at the NGEE Arctic study sites on the Seward Peninsula, Alaska. Our ELM improvements include the use of local parameterizations using site-specific data sets, subgrid topographic model frameworks, and model structure enhancements for shrub allometry. Our work is important for the development of improved Earth system models that can capture snow-vegetation-permafrost interactions and better characterize the Arctic carbon balance as a whole, which is the focus of the next phase of the work currently underway.
  [poster]

Aufeis, layered ice that forms through repeated cycles of flooding and freezing, is a critical water storage reservoir and seasonal habitat on river floodplains of Northern Alaska. Despite its ecological significance and sensitivity to environmental change, the long-term spatial and temporal patterns of aufeis remain poorly understood at regional scales due to the frequency of observations and challenges in differentiating snow and ice in optical imagery. In this study, we developed an automated pipeline using the Google Earth Engine Python API to detect and analyze aufeis across the North Slope Borough from 1984 to 2024. We applied a random forest classification model to over 5,000 multispectral Landsat and Sentinel-2 images, supported by geographic constraints, to map the maximum area of aufeis and its decay between May 15 and September 15. Logistic regression models were fit to each pixel based on aufeis presence or absence and used to assess long-term trends and intra-seasonal dynamics in aufeis occurrence. Our results reveal significant spatial heterogeneity in aufeis formation and changes over time, with both expanding and contracting trends observed across the North Slope. This study offers the first scalable analysis of aufeis across Alaska’s northern region and its response to Arctic change, highlighting evolving groundwater-surface water interactions and cryosphere processes in permafrost landscapes.
  [poster] 

Arctic and Boreal regions are important sources of methane hotspots. Given the role of methane as a potent greenhouse gas, understanding methane emission sources is a critical part of the global carbon cycle. Methane hotspots, which may have orders of magnitude higher methane emissions than background rates, have been detected in parts of Alaska and northwestern Canada using the Airborne Visible Infrared Imaging Spectrometer - Next Generation (AVIRIS-NG) as part of NASA’s Arctic Boreal Vulnerability Experiment (ABoVE). Previous research has found that methane hotspots correlate with thermokarst probability at regional scales and proximity to water bodies at local scales. In this investigation we tested the hypothesis that the methane hotspot correlations extend to individual permafrost disturbance features at the landscape scale. Using the Permafrost Region Disturbance (PRD) database, we correlated methane hotspots with landscape-scale permafrost disturbances and sub-surface degradation in transitional permafrost environments. The PRD database, which includes fire scars, lake loss, and thaw slumps, is based on a time-series analysis of 30-m resolution Landsat imagery from 1999 to 2014. We found that methane hotspots correlated with proximity to lakes with drying trends. We further computed the distances from methane hotspots to the nearest PRD feature(s) and quantified this information to understand spatial clustering and occurrence probabilities. Our research highlights the importance of landscape scale permafrost disturbances in understanding methane emissions in Arctic and Boreal regions.

My thesis evaluates lake sediment characteristics and how they have been affected as large geologic methane (CH4) seeps formed in Esieh Lake (informal name), a lake in Northwest Alaska. I provide extensive background of the lake, including a synthesis of studies and reports that characterize the geology around this lake which, to date, contains the largest known CH4 seeps in the Arctic. In addition to providing background information on Esieh Lake and characterizing the lake’s sediments, I evaluate the economic potential for the CH4 seeps and compare flux values to natural gas projects which were previously completed in Alaska and Canada. The evidence suggests the possibility that CH4 seeps initiated in Esieh Lake sometime within the last century via an explosive event that formed large pockmarks in the lake bottom. Rapid expansion of the seep field occurred between 1952 and 1972. Seepage continued after the blowout event and is still present today, albeit at a more quiescent stage. An economic evaluation of the seep as an energy source found that the capital cost for infrastructure to transport gas to a nearby community resulted in high energy costs, higher than the current cost of electricity in Noatak from imported diesel. However, if infrastructure capital costs were not a factor, then the cost of electricity for Noatak using Esieh Lake seepage as natural gas, would be much lower that current electricity costs. Through existing technologies, Esieh Lake is not economically viable as a resource but as technology progresses, developing a very small-scale gas resource may become a viable option.
  [poster] 

Complex nonlinear relationships exist between the permafrost thermal state, active layer thickness, and terrestrial carbon cycle dynamics. In Arctic and boreal Alaska, significant uncertainties characterize the spatiotemporal rate and magnitude of permafrost degradation and the permafrost carbon feedback, with increasing recognition of the importance of thawing mechanisms. The challenges of monitoring sub-surface phenomena with remote sensing technology further complicate the issue. There is an urgent need to understand how and to what extent thawing permafrost destabilizes the carbon balance in Alaska and to characterize the feedback involved. In this research, we use our artificial intelligence-driven model GeoCryoAI to quantify permafrost carbon dynamics in Alaska. The GeoCryoAI model uses a hybridized process-constrained ensemble learning framework to simultaneously ingest, scale, and analyze in situ measurements, remote sensing observations, and process-based modeling outputs with disparate spatiotemporal sampling and data densities. We evaluated prior naïve (a) persistence and (b) teacher forcing approaches relative to (c) time-delayed GeoCryoAI simulations, yielding the following error metrics (RMSE) for active layer thickness (ALT), methane (CH₄), and carbon dioxide (CO₂), respectively: 1.997, 1.327, 1.007 cm [1963–2022]; 0.884, 0.715, 0.694 nmol CH₄ km-2 month-1 [1994–2022]; 1.906, 0.697, 0.213 µmol CO₂ km-2 month-1 [1994–2022]. Our approach overcomes traditional model inefficiencies and resolves spatiotemporal disparities. GeoCryoAI captures abrupt and persistent changes while introducing a novel methodology for assimilating contemporaneous information at various scales. We describe GeoCryoAI, the methodology, our results, and plans for future applications.
  [poster] 

Approximately 8100 km2 of Alaska are leased to the oil and gas industry for exploration and extraction. According to industry estimates, subsurface expansion from these leases could cover up to 130.2 km2 per pad. As industrial oil extraction activities increase across the thawing Alaskan permafrost, impacts on the permafrost environment will include rapid thaw, increased hydrological flux, and the release of climate warming greenhouse gases. Here, we use remote sensing and field observations to provide a first-order comparison of the direct impacts to the permafrost tundra from oil well pads, and the long-term consequences of a legacy oil pads on the warming North Slope of Alaska. We find that oil well pads on the permafrost accelerate permafrost degradation and persist despite remediation.

As beavers follow favorable habitat expansion into Alaskan tundra regions, the landscape modifications they produce raise questions about the ecological impact of beaver engineering on permafrost-affected landscapes. The impoundment of water by beaver dams can have profound repercussions on permafrost, leading to increases in soil temperatures that drive thaw and, in ice-rich soils, subsidence. However, the impact of beavers on permafrost stability at regional scales has not been quantified. Here, we quantify subsidence surrounding beaver ponds and beaver-free ponds on Alaska’s Seward Peninsula using radar interferometry. This study expands previous local-scale analyses, aiming to enhance our ability to predict permafrost stability in heterogeneous conditions. The research area encompasses a 53 kilometer-wide swath of arctic tundra stretching from Nome northward, covering lowland and upland/mountainous terrain and geomorphology. We leverage high-resolution UAVSAR interferometric radar from 2019 to 2022 to estimate surface deformation at 3 m resolution. We map subsidence adjacent to ponds relative to the surrounding uplands. Using a GIS database of beaver pond site points compiled from satellite imagery and in situ observations, we compare subsidence near beaver-affected and beaver-unaffected ponds. Preliminary results reveal up to 10 cm of subsidence adjacent to water bodies within the study area. A statistical comparative analysis between beaver-affected and unaffected ponds as a function of distance from the shoreline will be employed to evaluate the differential impacts of beaver activity on permafrost stability and subsidence dynamics across a diverse tundra landscape.
  [poster]

Methane (CH₄) is a potent greenhouse gas emitted from Arctic lakes, often through ebullition (“bubbling”). These ebullition bubbles can originate from a variety of sources, depending on substrate age and type (i.e. microbial, thermogenic, mixed). Discrete bubbling sites, known as “seeps”, can vary in mechanical strength and gas composition. During winter, seeps manifest as ice-trapped bubbles (weak bubbling) and thinly-crusted or open holes (strong bubbling) that vary in spatial extent across lake ice sheets. Previous work has shown that both small and large CH₄ seeps appear in winter L-band (~24 cm wavelength) space-borne synthetic aperture radar (SAR) imagery as high-backscatter perennial anomalies within lake ice (Engram et al., 2013; Engram et al., 2020; Engram and Walter Anthony, 2024). From this, we developed a method using ALOS-1 PALSAR imagery to identify SAR-detected areas of potential seepage (SAPS) in lakes on a regional scale using previously groundtruthed large, highly-ebullitive seeps emitting radiocarbon-depleted CH₄ (aka. "superseeps”) (Tyler et al., in prep). Here, we present updated maps of SAPS for the Utqiaġvik, Atqasuk, and Noatak regions in Alaska and compare our results with regional fault data, limnicity, and field data. Additionally, we present our recently-published SAPS dataset for Fairbanks-area lakes (Tyler, Engram, and Walter Anthony, 2025). References: Engram, Melanie et al. (2013) “Synthetic aperture radar (SAR) backscatter response from methane ebullition bubbles trapped by thermokarst lake ice,” Canadian Journal of Remote Sensing, 38(6), pp. 667–682. Available at: https://doi.org/10.5589/m12-054. Engram, M. et al. (2020) “Remote sensing northern lake methane ebullition,” Nature Climate Change, 10(6), pp. 511–517. Available at: https://doi.org/10.1038/s41558-020-0762-8. Engram, M. and Anthony, K.W. (2024) “Synthetic aperture radar (SAR) detects large gas seeps in Alaska lakes,” Environmental Research Letters, 19(4), p. 044034. Available at: https://doi.org/10.1088/1748-9326/ad2b2a. Tyler, N., Engram, M.J. and Walter Anthony, K. (2025) “Potential Areas of Gas Seepage Detected by SAR in Lakes near Fairbanks, Alaska.” ORNL DAAC. Available at: https://doi.org/https://doi.org/10.3334/ORNLDAAC/2394

Ice-wedge systems (IWS) form polygonal ground that is ubiquitous throughout the Arctic. Processes of ground ice accumulation and degradation drive the diversity of IWS polygon types: low-centered polygons (LCP) have centers below elevated rims that were formed from displacing soil during wedge ice development, while high-centered polygons (HCP) typically represent stages of partial degradation as subsidence in troughs caused from melting of the top of the wedge ice causes the centers to become elevated. Recent remote sensing observations have shown rapid change of LCPs transitioning to HCPs, however knowledge gaps remain that quantify the environmental response of IWS evolution. We present results monitoring two 100 m by 100 m grids in 2023 and 2024, one containing LCPs and the other HCPs, at the Teshekpuk Lake Observatory in northern Alaska. We collected point measurements every 10 m of thaw depths, ground temperatures (at 12 cm depth), soil moisture (or water depth), and maximum vegetation heights in July 2023 and 2024 and snow depths in May 2024. UAV surveys of the grid sites mapped micro-topography and peak-summer NDVI. Furthermore, ground temperature loggers were placed in specific geomorphic features (polygon centers, troughs and rims) to record active layer (15 cm depth) and permafrost (100 cm depth) temperatures throughout the year. Our results show that the shift from LCP to HCP leads to a divergence in environmental responses including shifting vegetation productivity, soil drying and ground cooling. We found that mean conditions varied less between both sites (e.g., NDVI: +15.4%, thaw depths: +7.3%, ground temperature: -4.9 %), while standard deviations values varied more overall (e.g., NDVI: +35.3%, thaw depths: +38.5%, ground temperature: +21.1%). We interpret data on ends of the data distribution and outliers of environmental parameters represent areas of ice-wedge degradation, therefore remote sensing and modeling studies relying on coarser products are not adequately capturing the finer-scale processes of ice-wedge heterogeneity.

The soil organic carbon (SOC) accumulated in the Arctic permafrost and its active layer accounts for the largest terrestrial carbon pool on Earth. In the past few decades, anthropogenically induced warming has triggered the Arctic amplification, referring to the warming trends in the Arctic and northern high latitudes being faster than any other place on Earth. Currently, there is a large uncertainty in the response of these vast SOC reservoirs to the warming trends in the permafrost regions. Recently a meta-analysis among Arctic ecosystem modelers indicated the quantification of SOC (both stock and dynamics) as the major data gap and contributor to this uncertainty. In response, we are undertaking a community synthesis effort under the NASA Arctic Boreal Vulnerability Experiment (ABoVE) focusing on a regional consolidation and harmonization of soil properties to support science, modeling, and remote sensing efforts within boreal-Arctic Alaska. While initially we aim to utilize the synthesized SOC to assist in validating the radar-derived SOC estimates collected as part of ABoVE Airborne Campaign, the synthesized SOC dataset allows pursuing other research interests such as investigating the SOC response with respect to fire disturbances in the tundra and the retrogressive thaw slump expansions. We have spent the last few months synthesizing in-situ SOC data from the previous three decades in the Tundra region of Alaska. In order to more effectively utilize the data that have been collected over the ABoVE decadal time frame and beyond, we are seeking to harmonize, standardize, and publish the data on one consolidated platform. We have communicated with other PIs who collected the individual data sets (66 PIs who own 88 datasets), as well as other researchers who may be interested in contributing to this synthesis effort. This collaboration spans many interests, including permafrost, carbon, hydrology, and climate. The goal is to archive the data (1400 cores, 15000 soil samples) in one seamless location on ORNL DAAC, which will potentially reduce the uncertainties in regional SOC estimates, and thus create a synthesis paper to be created throughout the summer of 2025. We further hope that impactful science can be conducted from the synthesized data for an improved understanding and improved trend projections of the Arctic ecosystem.
  [poster]

For this study, we used spaceborne lidar-derived vegetation observations and disturbance data from Landsat’s long historical record to examine the spatial patterns of forest growth across the circumpolar boreal region. By combining estimates of forest height from ICESat-2’s ATL08 land and vegetation data product with forest age from Landsat we are able to estimate site index, a measure of forest growth potential, using spaceborne remote sensing data - critical for regions like the circumpolar boreal, which is largely comprised of remote areas with limited ground measurements. Our analysis reveals significant spatial variations in forest growth patterns, with observable trends that may indicate ecosystem shifts in response to warming temperatures. We found that Russia, accounting for 73% of the estimated forest change, exhibits significant departures from the modeled expected growth rates of 2.4 to 4.8 meters ( ±?3.8?m) at the 75th and 90th percentiles. These findings suggest that, if allowed to recover from disturbances, young forests in Russia could serve as a substantial carbon sink. This work helps enhance our understanding of circumpolar boreal carbon dynamics and demonstrates new remote sensing methodologies for monitoring regional-scale forest productivity.
  [poster]

One of the objectives of the Phase3 of our ABoVE project is to quantify Above Ground Biomass (AGB) change across environmental gradients in Boreal forests. To this goal, we incorporated Sentinel1 C-band SAR into our AGB product workflow to enhance the model, reduce noise, and increase the model sensitivity across structure gradients. Since AGB is directly estimated from canopy height metrics, we focused on studying the effects of SAR in the quality of our canopy height maps (CHM) by conducting two large scale experiments. First is a comparison of our CHM against an independent validation set of LVIS transects conducted in the years 2017 and 2019. This dataset includes gridded relative height and canopy cover metrics for over 85 million, 30 meter squared pixels in the ABoVE domain. The second is a cross validation of our CHM against the 98th percentile of Icesat2 ATL08 relative height metric in the full circumpolar boreal domain. In both cases, we examine the residuals across a range of slopes, land cover classes, reference heights and canopy cover fractions. Our results highlight the importance of C-band SAR in relatively flat areas of the Boreal forests where land is mostly covered by short sparse trees, shrubs, wetlands, and herbaceous vegetation.
  [poster] 

While canopy chlorophyll content (Chl) informs on the potential for photosynthetic function, solar induced chlorophyll fluorescence (SIF) offers a directly probe to assess vegetation photosynthesis at leaf, canopy, and up to global scales. Recent technology developments are enabling continuous diurnal and seasonal measurements of leaf chlorophyll fluorescence, and canopy SIF and spectral reflectance for extended periods. Time series of these data were collected at canopy level using tower-mounted FLoX (Fluorescence boX, JB Hyperspectral) and at leaf level using Moni-PAM (Monitoring Pulsed Amplitude Modulation, Walz GmbH) sensors for tundra at Utqiagvik, AK and a boreal forest at Caribou Creek, AK, in conjunction with eddy covariance measurements of gross primary productivity (GPP). We compare the trends and findings from the analysis of these data collections and will discuss the observed differences and similarities.

The work described here has been part of a multi-year effort between Michigan Tech Research Institute, the University of Massachusetts, JPL, Michigan Tech University (Evan Kane), UCAR (Adriana Foster, Jackie Shuman, now at NASA-Ames), U. Geulph, Wilfred Laurier Univ., Natural Resources Canada, Govt. of NWT, and others for integrating remote sensing and modeling to better undestand the vulnerability of the ABoVE region ecosystems to wildfire. In the efffort described here we focus on the remote sensing aspect of the analysis, where C- and L-band SAR were used for characterizing wetlands and biomass distribution in the Arctic-Boreal region, with an emphasis on learning how satellite resources of Sentinel-1 and the upcoming NISAR mission can be used for creating time-series that would be used to measure the status and state of change for this hydrologically dynamic region.
  [poster]

Trends in peak spectral greenness have long been used to characterize changes in productivity in the Arctic tundra in response to global climate change. Yet spectral greening does not incorporate potential changes in phenological timing and growing season length, which respond uniquely to environmental conditions and have major implications for carbon storage, surface energy balance, and trophic interactions. While field-based studies indicate strong temperature sensitivity of green-up timing, snowmelt timing also likely exerts direct control, and there is less consensus on the controls constraining green-down timing. Remote sensing based studies have also found temperature dependency for spring green-up, but tend to disagree on the magnitude (and even the sign) of long-term trends in senescence timing and growing season length, with results varying by sensor and modeling techniques. Thus a need exists to better assess LSP from remote sensing across spatial scales for heterogeneous Arctic vegetation, and to compare results with field-based measurements. To address this challenge and unanswered questions linking land surface phenology (LSP) with environmental controls in the Arctic, we examined the following questions: (1) How has the shape of the phenological curve and in-turn the period of carbon uptake changed across the Arctic over the last two decades, (2) How are spring warming and snowmelt, summer warming, and hydrological conditions controlling LSP patterns? And (3) How do LSP and patterns related to environmental controls compare across spatial scales? To estimate LSP, we used cubic splines fit to time series of remotely sensed vegetation indices, fine-tuning them for the Arctic domain. Trend analyses were conducted with MODIS NBAR (500 m; MCD43A4) time series from 2001-2023. Climate sensitivity was analyzed using MODIS NBAR and Harmonized Landsat Sentinel-2 (30 m; HLS) time series from 2016-2023, with climate data from the ERA5-Land hourly data set (0.1°). LSP results were also compared to estimates from PhenoCam sites in Alaska. Preliminary results showed strong agreement between LSP retrieved from in-situ PhenoCams and estimated from HLS and MODIS. We found that the timing of green-up is highly sensitive to spring temperatures across all ecoregions. The timing of green-down is less variable and more constrained by photoperiod but appears to vary as a function of anomalies in pre- and post-solstice temperatures. We also found a geographically heterogenous increase in growing season length of about 1 day/decade since 2001, with a majority of pixels showing significant trends experiencing earlier and amplified vegetation growth (~64%), followed by universal growth throughout the growing season (~28%).
 

Fine-(meter) scale vegetation height change is ecologically significant, difficult to resolve from space, and occurs in a heterogeneous manner across broad spatial extents. There are many global canopy height models (CHMs) gemerated from machine and deep learning. These efforts leverage large volumes of data and optimize for global coverage over regional precision/accuracy. For shrubs and trees in the high northern latitudes (heights ~0.25m-15 m), which require rigorous precision to resolve a height change signal within a spaceborne time-series, this trade-off results in height estimates with large uncertainties for short stature woody vegetation. Here, we evaluate deep learning vegetation height prediction from commercial multi-spectral HR surface reflectance using two approaches: A UNet regression neural network; and fine-tuning of existing Vision Transformer (ViT) model (DINOv2) for remote sensing (RS) data of boreal vegetation. Results are assessed based on reference data collected on the Seward Peninsula, Alaska during summer 2024.

The Alaska Geospatial Council Vegetation Technical Working Group (AGC VTWG) is an advisory body consisting of multiple agencies and organizations that provides guidance on development of geospatial data related to vegetation for the state of Alaska. One overarching need they identified is to improve the Consistency in vegetation classification and mapping standards so vegetation inventory and mapping products may be integrated to develop statewide and cross-border products. Building on and continuing some of the collaborations, datasets, and innovative methods developed as part of the NASA Arctic Boreal Vulnerability Experiment (ABoVE) campaign, members of the VTWG are producing AKVEG Map, a repeatable mapping framework that is flexible to multiple user needs, promotes data compatibility among different agencies and protocols, provides high spatial and thematic resolution, and enables agencies to benefit from the sampling investments of others regardless of jurisdictional boundaries. Components include the U.S. National Vegetation Classification, a harmonized plot database, and a suite of geospatial products.

Retrogressive Thaw Slumps (RTS) are increasingly transforming Arctic permafrost landscapes, leading to significant shifts in vegetation composition and carbon cycling. However, large-scale assessment of vegetation change across RTS chronosequences is constrained by the limited availability of high-resolution hyperspectral data. In this study, we evaluate the potential of hyperspectral image simulation for mapping fractional cover (fCover) of dominant plant functional types (PFTs) across RTS features in the Noatak National Preserve, Alaska. We leverage a multi-scale dataset integrating ground-based vegetation surveys, high-resolution airborne imagery (RedEdge-M at 5 cm and AVIRIS-3), and Sentinel-2 MSI satellite data. Hyperspectral reflectance was simulated from Sentinel-2 using the Universal Pattern Decomposition Method (UPDM), and PFT fCover was estimated using a reference classification map. Comparative analysis using data from July 2023 demonstrated that simulated AVIRIS-3 imagery improved fCover estimation accuracy, with an average 8.9% increase in correlation and a 16.6% reduction in RMSE compared to Sentinel-2. The model was subsequently applied to assess growing season vegetation dynamics for 2022 and 2023 (May–September). Now we aim to extend this framework to produce vegetation maps over RTS scars—both manually delineated and identified via deep learning models throughout the ABoVE domain. Using representative endmembers based on Arctic bioclimatic zones to generate images using Sentinel-2 and UAS data. We will investigate vegetation succession across diverse tundra types (e.g., wetlands, graminoid, prostrate-shrub, erect-shrub, and barren). This approach enables a spatially explicit analysis of vegetation recovery dynamics, improving our understanding of ecosystem responses to thaw-related disturbances across heterogeneous Arctic climates and environments.
  [poster] 

Beavers (Castor canadensis) have rapidly expanded their ranges into tundra ecosystems of the Arctic in recent decades, and it is becoming clear that beaver engineering activities drive major geophysical change in these areas underlain by permafrost. It remains unclear, however, how beaver engineering - specifically the ponds they create that increase surface water extent and thaw surrounding riparian permafrost - may impact Arctic ecological communities. One of the most important components to the change may be in the vegetation community, which responds quickly to flooding as well as permafrost thaw that may extend up to 60m away from a new pond edge. This study investigates the impacts of beaver ponding on vegetation dynamics using plant functional traits (PFTs) derived from AVIRIS imagery from northwest Alaska. 469 beaver ponds and 171 control streams (areas unimpacted by beaver ponds) were identified across 2019 AVIRIS footprints, and average PFTs were compared at 10m increments away from pond or stream edges. Preliminary results indicate an increase in vegetation productivity (NDVI) between 10 and 30m away from beaver ponds, but no differences in other metrics of photosynthetic activity or plant health were found. This poster presents recent updates on this ongoing research project.
  [poster] 

This study evaluated deep-learning models Convolutional Neural Network (CNN), ResNet50, VGG19, U-Net, and Vision Transformer (ViT) for detecting changes in tall shrub cover across Alaska’s North Slope using very-high-resolution QuickBird (QB) and WorldView (WV) satellite imagery spanning 2002 to 2020. Models were trained on 4,100 image tiles (200-by-200 m) using various spectral inputs including Pansharpened 4-band multispectral satellite imagery and panchromatic (P1BS) at top-of-atmosphere (TOA) radiance. Model validation against the Toolik Lake Vegetation Community Map used as field-reference indicated robust accuracy across all architectures, with ResNet50 and VGG19 achieving highest accuracies (~86%) and balanced F1 scores (57–59%). CNN and U-Net models also performed moderate (81–85% accuracy), with CNN exhibiting effective hierarchical feature extraction. Supplementary inputs from stacked imagery, such as the Normalized Difference Vegetation Index (NDVI) and Principal Component Analysis (PCA) scenes, did not substantially improve segmentation accuracy, suggesting that linear transformed TOA radiance imagery products do not further improve model performance. Binary segmentation predictions revealed subtle variations among models, particularly in transitional vegetation zones. Continuous tall shrub cover predictions showed that model performance was negatively correlated with shrub density (r = -0.85 for accuracy; r = -0.54 for F1), implying greater classification challenges in denser shrub areas. Additionally, one-way ANOVA tests indicated significant, but modest sensor-driven differences (QB02, WV02, WV03) in shrub cover estimates. Temporal trend analyses demonstrated a statistically significant increase in tall shrub cover over time for all models, with the ResNet50 models showing consistently highest positive slopes (?0.86% per year, p<0.001) while the CNN, VGG19, U-Net and ViT models showed more modest growth rates (0.08%, 0.29%, 0.25% and 0.04% per year respectively, p<0.001). Binary segmentation thresholding methods influenced estimated shrub expansion rates, highlighting that dynamic percentile-based thresholds (p70–p80%) produced conservative and reliable (tall) shrub delineations compared to global thresholds. Calibration against field-reference indicated strong predictive capability for ResNet50 and VGG19 models (R²=0.78–0.81), with relatively low errors and biases. Comparisons with NASA’s G-LiHT-derived Canopy Height Model (CHM) further validated model robustness, revealing moderate correlations (R²=0.18–0.27) between model-predicted shrub fractions and CHM-derived canopy heights. Cross-sensor comparisons with Planet imagery, utilizing the same architectures (VGG19, ResNet50), yielded broadly consistent spatial shrub patterns, though Planet-derived predictions exhibited higher interannual variability due to lower resolution and calibration differences. Despite these sensor-specific variations, both Planet and QB/WV datasets demonstrated consistent temporal trends in shrub expansion. This study highlights the efficacy and reliability of deep-learning methodologies for mapping and monitoring tall shrub cover across heterogeneous Arctic landscapes. Our findings underscore the importance of sensor calibration, appropriate spectral inputs, and robust segmentation strategies in accurately detecting shrub dynamics. Continuous expansion of tall shrub cover aligns with documented Arctic warming trends, emphasizing the potential for deep-learning-based approaches to advance our understanding of permafrost dynamics and vegetation-climate feedbacks in rapidly changing northern ecosystems.
  [poster]

As NASA's Arctic Boreal Vulnerability Experiment (ABoVE) nears the conclusion of its 10 plus year research program, understanding biomass estimation variability between datasets is critical for assessing carbon dynamics and disturbances in high-latitude ecosystems, which store a significant portion of global carbon. Here we have conducted a meta-analysis of eight biomass estimation datasets to evaluate their variability across the North American Arctic and Boreal Region (ABR). Biomass estimates were validated against ground plot data, compared at province and ecoregion levels, and assessed under different disturbance conditions including wildfire and timber harvest. Significant differences in dataset coverage, resolution, and methodologies led to variability in biomass estimates and therefore potential applications. This work provides key insights into biomass modeling approaches and datasets within the ABR, a region experiencing rapid climate change and shifts from carbon sinks to carbon sources. It supports NASA’s mission to understand the vulnerability and resilience of these at-risk ecosystems and their impact on societies.

A fundamental property of lakes is their connection to their watersheds. Inflow, vegetation, a complex shape, or relatively small surface area can all increase a lake’s connectivity to its watershed, which provides carbon and nutrient inputs. In changing Arctic landscapes, it is possible that highly landscape-connected lakes will be most sensitive to terrestrial trends and disturbances. Our work seeks to identify these lakes through a novel remotely sensed connectivity index derived from lake morphometry (i.e. shape and area) and aquatic vegetation properties (i.e. NDVI, greenness, coverage). The index is compared against geochemistry measurements from the Harmonized Arctic-Boreal Lake (HABL) database, a spatially extensive compendium of inland water carbon chemistry and flux from over 3,000 Northern high latitude lakes. We expect the connectivity index will correspond to indicators of terrestrially sourced organic carbon. With this relationship in place, it will be possible to identify across the ABoVE region lakes with strong terrestrial carbon pools at risk of changing water quality regimes. Preliminary findings show that lake greening and browning trends in both vegetation and water color are decoupled from terrestrial trends. Even so, aquatic vegetation ranges are expanding north, similar to woody vegetation. Our results will help understand why certain lakes respond more to landscape change than others.
  [poster]

For ABoVE Phase 3-funded research, we have developed the Harmonized Arctic-Boreal Lake (HABL) database to compile field and remote sensing measurements of inland water carbon chemistry and flux at over ~3000 northern high latitude lakes. The HABL provides both a database and an interactive visualization platform to analyze historical in situ measurements of carbon concentration (DOC, DIC, CO2, CH4), composition (specific UV absorbance, fDOM, CDOM, BDOC, carbon and oxygen isotopes, FT-ICR-MS information), and flux (pelagic and littoral fluxes of CO2 and CH4) in conjunction with remotely sensed, physiographic, and climatic variables (e.g. optical properties, lake area, perimeter, land cover, permafrost probability, mean annual air temperature and precipitation, and drainage basin characteristics). This work represents the development of a major integrated and spatially explicit repository of lake chemistry data for northern latitudes and links datasets to investigate how (1) changing precipitation patterns and temperature influence observed concentrations of inorganic and organic carbon; (2) temporal heterogeneity of inundation extent drives variation in the concentrations and form of organic carbon and GHG emissions in lake ecosystems; and (3) climate, water surface area, and land cover drive trends in lake reflectance (proxies for productivity and carbon source). For each lake, shoreline extents were derived from existing inventories or semi-automated image classification. Using radar and optical satellite imagery, we quantify littoral vegetation and inundation extent and characterize the temporal trends in optical characteristics associated with open water at locations for which carbon chemistry and flux measurements are available. Through GIS spatial analysis, we associate watershed characteristics, such as soil organic carbon content and active layer thickness, with in-situ lake measurements, to enable coupled investigations of terrestrial and aquatic biogeochemical change. Further, using Google Earth Engine, we have built a community tool for users to upload their field data to HABL. Additional field datasets from the community are welcomed as HABL approaches the final stage of development, with anticipated public release in summer 2025.
  [poster] 

Northern lakes, especially lakes impacted by thawing permafrost, are a major source of atmospheric methane, a potent greenhouse gas. However, challenges in upscaling field data, including fluxes by ebullition (bubbling), lead to large uncertainties in emissions estimates. Previous work demonstrated that ebullition increases synthetic aperture radar (SAR) backscatter in the L-band frequency from ice-covered lakes by increasing the roughness of the ice/water interface (Engram et al. 2012, 2013). An empirical regression model using decomposed quadrature-polarized (quad-pol) SAR was created based on year-round flux data from underwater bubble traps, then validated in the Barrow Pen. and Atqasuk regions with independent airborne eddy covariance measurements (Engram et al. 2020). This work established L-band SAR as a useful tool to estimate gas ebullition from lakes that freeze and created the potential for detailed mapping of regional emissions on a landscape scale. However, too few quad-pol L-band SAR images were acquired to apply this analysis to new regions, creating a data availability problem. Here, we developed a novel regression model for predicting ebullition in ice-covered lakes using single-polarized (single-pol) L-band SAR data. Single-pol data are much more widely available and represent a platform analogous to the NISAR satellite data expected to be routinely provided later this year (2025). Our new L-band single pol regression model utilizes field data of annual ebullition from study lakes from five Alaska regions: Utqia?vik (Barrow Peninsula), Atqasuk, Toolik, northern Seward Peninsula, and Fairbanks. Data were acquired from the Phased Array type L-band Synthetic Aperture Radar (PALSAR-1) instrument on the Japanese Advanced Land Observing Satellite (ALOS-1) at ~39° incidence angle with a horizontal transmit and receive polarization (HH). Field data on lake ebullition were based on a combination of on-ice bubble surveys from 2007-2017 and semi-automated underwater bubble-traps per Engram et al. (2020). Single-pol (HH) L-band SAR backscatter was highly correlated with ebullition field data, consistent with our previous results using the T11 “roughness” component of quad-pol L-band SAR (Engram el al. 2020). In addition to the polarization difference, several differences between our previous T11 quad-pol and the new single-pol analyses should be noted. The quad-pol data were acquired with a steeper incidence angle (~24°) than the Fine Beam Single-pol data (~39°). Also, the pixel size of the quad-pol data was larger (12.5 m) than that of the Fine Beam Single-pol data (6.25 m). To examine the effect of SAR data variables isolated from polarization differences, we used lakes in the Fairbanks area which exhibit a wide range of gas fluxes. We compared only the HH channel of the quad-pol data for the single early winter scene to the sum of 13-17 winter HH Fine Beam Single-pol scenes for five years to examine the effects of incidence angle and more frequent temporal sampling later in the winter on efficacy of emission prediction. New L-band data from the upcoming NISAR mission coupled with such a single-pol SAR empirical regression model could provide current methane ebullition estimates from lakes that freeze.

Beavers are known for their role in enhancing ecosystem methane emissions in temperate and subboreal ecosystems by engineering flooded aquatic habitats and concentrating carbon stocks in anaerobic settings (Bubier 1993). While Clark (2024) proposed a relationship between hyperspectral AVIRIS-NG CH4 hotspots and beaver expansion into NW Alaska tundra, to our knowledge, observational field data of beaver-associated methane emissions are lacking for permafrost ecosystems. Here we conducted field measurements (n = 813) of diffusive methane fluxes from 14 beaver-impacted lakes, ponds and streams in Alaskan tundra and boreal forest ecotypes characterized by permafrost. Methane emissions within 3-m of beaver lodges were 5 to 38 times higher than emissions away from lodges. Summer emissions were more than twice as high as winter emissions from the same sites. The distribution of emissions among our 14 tundra and boreal sites was similar to that of the eight beaver-impacted sites at lower latitudes with sporadic or no permafrost reported in previous global methane inventories (Wik 2016; Kuhn 2021). Among our study sites, emissions were highest from thermokarst lakes and ponds formed within recent decades in Yedoma permafrost. Emissions were lower from sites where the surrounding permafrost soil carbon stocks were also lower. While this between-site variability appears to be driven by the same factors that control non-beaver-impacted aquatic methane emissions, such as soil and vegetation carbon stocks (Walter Anthony 2016), our data support the suggestion that beavers have the potential to increase ecosystem methane emissions at the local scale (i.e. dams, lodges) by concentrating organic carbon substrates, and at the regional scale by increasing inundation and thermokarst. Next steps of this research include a) expanding the geographic scope of fieldwork, b) incorporating field- and remote-sensing based estimates of ebullition, and c) comparing field flux observations to hyperspectral AVIRIS-NG CH4 hotspots mapping in the same regions. Bubier, J., T. R. Moore, N. T. Roulet. Methane Emissions from Wetlands in the Midboreal Region of Northern Ontario, Canada, Ecology 74, 2240-2254, https://doi.org/10.2307/1939577, 1993. Clark, J. A. et al. Do beaver ponds increase methane emissions along Arctic tundra streams? ERL, Environ. Res. Lett. 18, 075004DOI 10.1088/1748-9326/acde8e, 2024. Kuhn, M. A. et al. BAWLD-CH 4: a comprehensive dataset of methane fluxes from boreal and arctic ecosystems. Earth System Science Data 13, 5151–5189, 2021. Walter Anthony, K. M. et al.. Methane emissions proportional to permafrost carbon thawed in Arctic lakes since the 1950s. Nature Geoscience 9, 679–682, doi:10.1038/ngeo2795, 2016 Wik, M., Varner, R. K., Anthony, K. W., MacIntyre, S. & Bastviken, D. Climate-sensitive northern lakes and ponds are critical components of methane release. Nat. Geosci. 9, 99–105, 2016.
  [poster] 

Winter methane (CH4) emissions from frozen lakes represent an important gap in the boreal-arctic methane flux dataset (Kuhn 2021). Very few methane flux measurements have been collected in this region over the shoulder seasons and winter, and lake methane budgets at best include only emission terms for water column and ice-sheet bubble storage that can be released upon icesheet melt in spring (Kuhn 2021). Despite suggestions that winter season emissions could be significant (Karlsson 2013, Sepulveda Jauregui 2015), the global budget considers emissions from ice-covered lakes to be zero (Saunois 2024, Johnson 2022). We collected >2000 portable chamber diffusive flux measurements on 55 boreal and arctic lakes in Alaska during the winter ice-cover period and seven years of continuous eddy covariance flux measurements in an interior Alaska thermokarst lake, Big Trail Lake, from 2017-2025. Mean winter flux measurements were positive for 50 out of 55 lakes, ranging from 0.05 to 892 mg CH4 m-2 d-1. At Big Trail Lake, where mean daily winter emissions measured by eddy covariance were 117 ± 63 mg CH4 m-2 d-1, the ice cover season comprised 40% to 55% of total annual emissions. A first-order upscaling, taking into account lake type and size according to the aquatic classes of the Boreal–Arctic Wetland and Lake Methane Dataset (BAWLD-CH4) (Kuhn 2021) indicates that including winter diffusive methane fluxes will increase the estimate of total annual emissions from boreal and arctic lakes by 37%. Karlsson, J., Giesler, R., Persson, J., and Lundin, E.: High emission of carbon dioxide and methane during ice thaw in high latitude lakes, Geophys. Res. Lett., 40, 1123–1127, https://doi.org/10.1002/grl.50152, 2013.? Kuhn, M. A. et al. BAWLD-CH 4: a comprehensive dataset of methane fluxes from boreal and arctic ecosystems. Earth System Science Data 13, 5151–5189, 2021. Matthews, E., Johnson, M. S., Genovese, V., Du, J., and Bastviken, D.: Methane flux from high latitude lakes: methane-centric lake classification and satellite-driven annual cycle of fluxes, Sci. Rep.-UK, 10, 1–9, https://doi.org/10.1038/s41598-020-68246-1, 2020.? Saunois, M. et al. The global methane budget 2000--2020. Earth system science data, in press, doi.org/10.5194/essd-2024-115, 2024. Sepulveda-Jauregui, A., Walter Anthony, K. M., Martinez-Cruz, K., Greene, S., and Thalasso, F.: Methane and carbon dioxide emissions from 40 lakes along a north–south latitudinal transect in Alaska, Biogeosciences, 12, 3197–3223, https://doi.org/10.5194/bg-12-3197-2015, 2015. Wik, M., Varner, R. K., Anthony, K. W., MacIntyre, S. & Bastviken, D. Climate-sensitive northern lakes and ponds are critical components of methane release. Nat. Geosci. 9, 99–105, 2016.

The Arctic-Boreal region has experienced abnormally rapid warming and significant land cover changes (LCCs). Revealing the climatic effects of LCC is crucial for better understanding land-atmosphere interactions and constructing feasible land-based solutions for mitigating climate risks, especially in a vulnerable part of the world, such as the ABoVE domain. However, there is little study on investigating the biogeophysical effects of LCC on regional climate considering real-world land cover dynamics in the Arctic-Boreal region. This study combined realistic remotely-sensed land cover maps and the Community Earth System Model version 2 (CESM2) to examine how LCC affected regional mean climate as well as climate extremes in the ABoVE domain. The results revealed that LCC, particularly the net loss of forests, in the study region, has significantly altered the local climate from 1984 to 2014. The statistically significant alterations in albedo and net radiation resulted in a local cooling effect that surpassed 1 °C during spring. This also led to a statistically significant variation in temperature extremes, with ice days (ID) and warm spell duration (WSDI) increased by more than 3 days and maxima of daily maximum temperature (TXx) decreased by more than 0.6 ?. Changes in precipitation extremes were also statistically significant because of the combined effects of water vapor supply and atmospheric dynamics. The number of wet days (R10mm) and total precipitation on wet days (PRCPTOT) decreased by more than 3 days and 15 mm, respectively, while the maximum length of dry spell (CDD) increased by more than 3 days. These findings fill an important gap in the existing research on LCC and climate variability in the Arctic-Boreal region.
  [poster]

As northern high latitudes warm faster than any other region on Earth, they are undergoing major functional shifts with uncertain consequences for global carbon, water, and energy cycling. Here, we show how satellite and ancillary observations of terrestrial water storage (GRACE, GRACE-FO), snow cover fraction (MODIS), vegetation leaf area, photosynthesis and biomass can be assimilated into a process-based terrestrial biosphere model to provide a window into critical dynamical processes and interactions between the land biosphere’s carbon, water and energy cycles. We describe recent advances in a Bayesian model–data fusion framework (CARDAMOM), its underlying model (DALECCWE), and the framework’s implementation across the ABoVE domain region. We link these advances with paths forward for tracking permafrost melt, shifting streamflow temperature regimes, riverine dissolved organic carbon fluxes into coastal ecosystems, and shifts in aerobic and anaerobic microbial activity in soils, and demonstrate how Bayesian model-data integration is a key step towards resolving critical processes that cannot be directly observed.

Methane (CH₄) flux estimates from high-latitude North American wetlands remain highly uncertain in magnitude, seasonality, and spatial distribution. In this study, we evaluate a decade (2007 – 2017) of CH₄ flux estimates by comparing 16 process-based models from a recent inter-comparison project with atmospheric CH₄ observations collected from in-situ atmospheric observation towers across Canada and the US. We first compare the current Global Carbon Project (GCP) process-based models with a model inter-comparison from a decade earlier called WETCHIMP. Our analysis reveals that the current process-based models have a much smaller inter-model uncertainty and have an average magnitude that is a factor of two smaller across Canada and Alaska. With that said, the current GCP models likely overestimate wetland fluxes by a factor of two or more across Canada and Alaska based on our analysis using atmospheric CH₄ observations. Furthermore, the differences in flux magnitudes among GCP models are more likely driven by spatial-temporal uncertainties in soil carbon or inundation uncertainties than in temperature relationships, such as Q10 factors. In addition, we find that models that agree best with atmospheric observations show peak CH₄ fluxes in July and August, while models that exhibit a different seasonal timing or a flat seasonal cycle show seasonal discrepancies compared with atmospheric CH₄ data. Furthermore, models that exhibit the best fit to atmospheric observation have a similar spatial distribution; these models concentrate fluxes near the Hudson Bay Lowlands (HBL). Overall, current, state-of-the-art process-based models are much more consistent with atmospheric observations than models from a decade ago, but our analysis shows that there are still numerous opportunities for improvement.
  [poster] 

The Arctic-Boreal Vulnerability Experiment (ABoVE) is a campaign from the NASA Terrestrial Ecology Program. ABoVE uses a of combination of field measurements, remote sensing, and modeling to understand the processes and interactions controlling the vulnerability and resilience of the region’s social-ecological systems. Currently there are over 125 projects and over 1000 people that are part of the ABoVE Science Team, with scientists spread over multiple nations and research institutions. Research, including an extensive airborne campaign, has been conducted throughout Alaska and northern Canada. Developing a coordinated campaign has been a team effort with collaboration at many levels.

The Arctic-Boreal Vulnerability Experiment is a NASA Terrestrial Ecology Program field campaign being conducted in Alaska and western Canada. This campaign has included field work as a central part of its research strategy for the past decade. Fieldwork presents inherent safety hazards (physical, psychological, and emotional), and often occurs in remote areas that can be difficult to access and can result in situations where team members could be vulnerable to assault, harassment, or hostile environments and encounters. Pre-fieldwork hazards assessments and appropriate safety training has been critical to ensure that researchers are well-prepared and feel comfortable in field settings. ABoVE’s training efforts recognize that safety training is not limited to operational safety, and that one can’t be safe when feeling frightened or intimidated by other people in a field team. The ABoVE team has found communication to be an important part of the field safety strategy, beginning early in the planning process both within and across teams and continuing throughout the field work. Additionally, utilizing both formal and informal strategies in conflict resolution and communication based on mutual respect have been central to ABoVE’s field safety strategy. Over the past decade, there have been over 1000 safety courses completed by researchers trained in safety protocols, varying from instruction in rafting down Arctic rivers in the summer to snow-machining across frozen landscapes in the winter, and including modules addressing aspects of personal safety. The commitment to safety has helped researchers gain new insights into the complex environmental changes occurring in this remote landscape and the training efforts aim to prepare individuals and teams for these risks.

NASA’s Harmonized Landsat and Sentinel-2 (HLS) project recently released a comprehensive suite of Vegetation Indices (VIs) derived from HLS V2.0 L30 and S30 datasets, originating from the Landsat 8/9 and Sentinel-2 A/B/C satellites, respectively. These indices are essential for a variety of vegetation-related applications that depend on dense time series data. However, to ensure reliable results, the consistency of these VIs must be evaluated. This study aims to assess the consistency of VIs derived from HLS V2.0 on a global scale. The evaluation utilizes same-day L30 and S30 image pairs from diverse landscapes around the world. Over 136 million cloud-free pixel pairs were randomly sampled to assess uncertainty. First, the normalized Root Mean Square Deviation (RMSDIQR) and R² for each VI derived from L30 and S30 was calculated. Second, potential factors contributing to discrepancies were then investigated. The analysis revealed that large view azimuth angle differences can cause a slight increase in discrepancies for some VIs. Similarly, a high solar zenith angle, which occurs during the winter season, can also result in increased discrepancies for some VIs. Finally, the impact of different aerosol levels on the consistency of cloud-free observations was examined. The best agreement between L30 and S30 VIs was observed at low-low or low-moderate (moderate-low) aerosol levels, whereas high aerosol levels led to notable discrepancies. Additionally, low discrepancies was often found within a range of VIs values, while extremely low or high VIs exhibited higher uncertainty. The study provides critical insights for researchers relying on these indices for accurate vegetation monitoring and analysis.
  [poster]

As a part of the NASA ABoVE Spectral Imaging Working (SIWG) group actives, we learned there are a vast number of unexplored dimensions within the AVIRIS-NG dataset. A useful tool for many would be a Foundation model trained on this large image dataset that finer-scaled models could be built upon, saving time and potentiallly improving results. The spatial and spectral information in these type of data provides an excellent opportunity to leverage the well-known strengths of Deep Learning (DL) extract patterns in images, in this across 3D scales (2D spatial, 1D spectral). To do this we built an Encoder/Decoder with a 3D Convolutional layers to Transformer (ViT) pipeline using the 2019 and 2019 AVIRIS-NG data on the ABoVE Science Cloud (ASC). We built the Encoder/Decoder with multiple Convolutional layers at different spatial (eg. 1x1 pixel, 3x3 pixel and up) and spectral scales (1nm, 5nm, etc.) to capture nuanced patterns in surface properties. The latent variables from Encoder this step were passed to the ViT, which uses fully connected Multi-head Attention blocks in series. To tune this model, we aim to use a range of validation locations with labels based UAS-flights conducted under AVIRS-NG by SIWG members. We expect that this Foundation model will be used for potentially understanding patterns in traits and fractional cover of vegetation interactions with soil, rock and litter with snow.