ALBERT

Description: Infrastructure and anthropogenic impacts are expanding across the Arctic. A consistent record of human impact is required in order to quantify the changes and to assess climate change impacts on the communities. We derived a first panarctic satellite-based record of expanding infrastructure and anthropogenic impacts along all permafrost affected coasts (100 km buffer) within the H2020 project Nunataryuk based on Sentinel-1/2 satellite imagery. C-band synthetic aperture radar and multi-spectral information is combined through a machine learning framework. Depending on region, we identified up to 50% more information (human presence) than in OpenStreetMap. The combination with satellite records on vegetation change (specifically NDVI from Landsat since 2000) allowed quantification of recent expansion of infrastructure. Most of the expanded human presence occurred in Russia related predominantly to oil/gas industry. The majority of areas with human presence will be subject to thaw by mid-21st century based on ground temperature trends derived from the ESA CCI+ Permafrost time series (1997-2019). Of specific concern in this context are also settlements located directly at permafrost affected coasts. An efficient erosion rate monitoring scheme needs to be developed and combined with settlement records in order to assess the risk for local communities and infrastructure. Relevant progress in the framework of the ESA EO4PAC project will be discussed.

Description: Retrogressive thaw slumps (RTS) are typical landforms indicating processes of rapid thawing and degrading permafrost. Their abundance is increasing in many regions and quantifying their dynamics is of high importance for assessing geomorphic, hydrologic, and biogeochemical impacts of climate change in the Arctic. Here we present a deep-learning (DL) based semantic segmentation framework to detect RTS, using high-resolution multi-spectral PlanetScope, topographic (ArcticDEM elevation and slope), and medium-resolution multi-temporal Landsat Trend data. We created a highly automated processing pipeline, which is designed to allow reproducible results and to be flexible for multiple input data types. The processing workflow is based on the pytorch deep-learning framework and includes a variety of different segmentation architectures (UNet, UNet++, DeepLabV3), backbones and includes common data transformation techniques such as augmentation or normalization. We tested (training, validation) our DL based model in six different regions of 100 to 300 km² size across Canada, and Siberia. We performed a regional cross-validation (5 regions training, 1 region validation) to test the spatial robustness and transferability of the algorithm. Furthermore, we tested different architectures, backbones and loss-functions to identify the best performing and most robust parameter sets. For training the models we created a database of manually digitized and validated RTS polygons. The resulting model performance varied strongly between different regions with maximum Intersection over Union (IoU) scores between 0.15 and 0.58. The strong regional variation emphasizes the need for sufficiently large training data, which is representative of the diversity of RTS types. However, the creation of good training data proved to be challenging due to the fuzzy definition and delineation of RTS. We are further continuing to improve the usability and the functionality to add further datasets and classes. We will show first results from the upscaling beyond small test areas towards large spatial clusters of extensive RTS presence e.g. Peel Plateau in NW Canada.

Description: With the Earth’s climate rapidly warming, the Arctic represents one of the most vulnerable regions to environmental change. These northern high latitude regions experience intensified fire seasons and especially tundra fires are projected to become more frequent and severe. Fires in permafrost regions have extensive impacts, including the initiation of thermokarst (rapid thaw of ice-rich ground), as they combust the upper organic soil layers which provide insulation to the permafrost below. Rapid permafrost thaw is, thus, often observable in fire scars in the first years post-disturbance. In polygonal ice-wedge landscapes, this becomes most prevalent through melting ice wedges and degrading troughs. The further these ice wedges degrade, the more troughs will likely connect and build an extensive hydrological network with changing patterns and degrees of connectivity that influences hydrology and runoff. While subsiding troughs over melting ice wedges may host new ponds, an increasing connectivity may also subsequently lead to more drainage of ponds, which in turn can limit further thaw and help stabilize the landscape. To quantify the changes in such dynamic landscapes over large regions, highly automated methods are needed that allow extracting information on the geomorphic state and changes over time of ice-wedge trough networks from remote sensing data. We developed a computer vision algorithm to automatically derive ice-wedge polygonal networks and the current microtopography of the degrading troughs from very high resolution, airborne laserscanning-based digital terrain models. We represent the networks as graphs (a concept from the computer sciences to describe complex networks) and apply methods from graph theory to describe and quantify hydrological network characteristics of the changing landscape. In fire scars, we especially observe rapidly growing networks and fast micromorphological change in those degrading troughs. In our study, we provide a space-for-time substitution comparing fire scars throughout the Alaskan tundra of up to 70 years since the fire disturbance, to show how this type of disturbed landscape evolves over time.

Description: Using our custom visualization tool for multitemporal Landsat satellite imagery we discovered, to our knowledge, an undocumented mega-landslide in far-east Siberia, which occurred in summer 2017 (https://bit.ly/2WYRLM1; 61.55°N; 170.01°E). To create and visualize this unique dataset, we processed temporal trends of multispectral indices of 〉100,000 Landsat images for a period from 2000-2019 using the freely available Google Earth Engine cloud processing platform (https://ingmarnitze.users.earthengine.app/view/hotspottcvisapp). The megaslide has a size of 3.66 km² and using the ArcticDEM data we estimate a volume movement of ~20 Mm³. With this size and volume, the landslide is among the largest globally known in recent decades. The landslide is accompanied by a smaller one (0.31 km², 1 Mm³) about 600 m further east, which already occurred in summer 2015. The large landslide caused the formation of several small lakes by blocking two valleys with debris and within newly formed crevasses near the hilltop, which are still persisting as of August 2021. As this event occurred in a remote valley far from any settlement, no visible damage to infrastructure or human livelihoods was detected. The remoteness has likely led to being not detected, like many similar, albeit a lot smaller, erosion features in the Arctic permafrost region. In this presentation we will show the main properties of the landslide, potential trigger mechanisms in the traditional sense. As this region is located along the fringes of permafrost presence we will discuss its potential connection to the rapidly warming climate in the high latitudes. Further, we will discuss how such a large event remained undetected for several years. We discuss and highlight the value of our landscape change visualization tool based on Landsat trend analysis (see Nitze et al., AGU 2020), which helped us to identify this extreme event. With more and more available data sources, this tool in addition to automated image analysis (e.g. deep-learning) or seismic analysis will help to uncover the hidden processes and dynamics of the Earth’s surface.

Description: Several decades of research have provided insight into patterns of and controls on thermokarst initiation and expansion, yet studies tend to focus on individual types of thermokarst (i.e., thaw lake formation and subsequent drainage) in particular regions. Today, we are left with uneven knowledge about abrupt permafrost thaw both conceptually and regionally. The goal of this presentation is to summarize recent advancements in monitoring thermokarst and its impact on soil, vegetation, and water while also framing a call to action for the next decade of research. Over the next decade, permafrost researchers must align their efforts on several fronts to not only increase our knowledge about changing permafrost but to align this knowledge with key community and policy needs. To support climate change planning and adaptation, northern communities need future thaw vulnerability mapped at scales relevant to their needs, which will require a suite of downscaled and new mapping and remote sensing products. Thermokarst predisposition maps based on circumpolar datasets greatly overestimate the area vulnerable to thermokarst, which can lead to poor planning and climate anxiety. In some situations, existing mapping products may be useful for downscaling with more detailed input data. In other situations, entirely new approaches may be required to support local action. A second key need for community relevant research is the ability to detect and monitor early warning indicators of thermokarst. Such information is needed to support scenario planning and to help mitigate the risks to social, cultural, and physical infrastructure created by permafrost change. We are evaluating the potential for using changes in vegetation, wetting/drying and topography as early warning indicators of thermokarst, all of which can be remotely sensed. Finally, integrating fine-scale disturbances such as thermokarst into large scale models remains a key challenge but critical for supporting sound climate policy. While a diversity of permafrost modeling approaches is necessary, we outline guiding principles that will help enhance model comparisons, assimilation of simulated data across spatiotemporal scales, and the ability for policy decisions to be rapidly informed by emerging science on permafrost change.

Description: Methane emissions from northern high latitude wetlands constitute a major uncertainty in the atmospheric methane (CH4) budget during the Holocene. To reconstruct northern wetland methane emissions, we used an empirical model based on syntheses of observations of peat initiation from more than 3600 radiocarbon-dated basal peat ages, plant-macrofossil-derived peatland type from more than 250 peat cores from sites across the northern high latitudes, and observed CH4 emissions averaged from modern-day wetland types in order to explore the effects of wetland expansion and changes in wetland type. Peatland basal ages and plant macrofossil records showed the widespread formation of fens in major northern wetland complexes before 8000 BP. After 8000 BP, new fen formation continued, but widespread peatland succession (to bogs) and permafrost aggradation also occurred. Reconstructed CH4 emissions from peatlands increased rapidly between 10,600 BP and 6900 BP due to fen formation and expansion, then stabilized after 5000 BP at 42 ± 25 Tg CH4 y-1, as high methane-emitting fens transitioned to lower methane-emitting bogs and permafrost peatlands. Permafrost formation in northern peatlands after 1000 BP decreased CH4 emissions by 20% to 34 ± 21 Tg y-1 by the present day. Warming temperatures, changes in peatland hydrology, and permafrost thaw will likely change the magnitude of northern peatland emissions in the future.