Permafrost and seasonally frozen ground affect large parts of the high latitudes and altitudes. The associated freeze/thaw cycles of the ground surface occurring during the transitional periods have been shown to be an important factor in, among others, the landscapes hydrology, it’s biological and ecological processes as well as greenhouse gas exchange. Freeze/thaw cycles are projected to undergo drastic changes in the coming decades due to climate change. Knowledge about the dynamics of the seasonality as well as the capability to monitor those processes will become even more important as changes progress. Active and passive microwave remote sensing sensors (e.g. C-band and L-band) have been used extensively to monitor Arctic landscapes. Applications include soil moisture retrieval, research on lake and sea ice detection and retrieval of the surface state of these landscapes. The accuracy of the resulting freeze/thaw data products is highly influenced by the landscape itself. It has been shown that different freeze/thaw data products based on various sensors can differ considerably in complex terrain (e.g. mountainous or lake rich areas). These deviations are often an issue of scale as the sensors differ significantly in both temporal and spatial resolution. It has been shown that number of frozen days, which can be derived from microwave remote sensing data, can be used to identify areas with potential permafrost beneath. Uncertainties are however high in transition zones and complex landscapes
In this study we investigate the possibilities and advantages that come with combining multiple sensors with different inherent sensor characteristics. We use scatterometer measurements from the Advanced Scatterometer (ASCAT) sensor on top of the MetOp satellites (C-band) as its temporal resolution (approx. daily measurements) allows us to capture short-term developments. However, the coarse spatial resolution of 25km (gridded to 12.5km) does not allow for an adequate understanding of the spatial patterns of these processes. Synthetic Aperture Radar (SAR) data from Sentinel-1 (C-band) and ALOS-2 PALSAR (L-band) has the advantage of a finer spatial resolution (approx. 40m) and therefor can be applied to understand within-footprint variability of freeze/thaw dynamics and compliment the coarse scatterometer data. Two study sites in northern Finland and the Austrian Alps were chosen, representing different landscapes forms. The study sites differ in their terrain, their vegetation and water body abundance. Additionally the alpine site exhibits glaciers and denser forest. For both study sites a network of near-surface ground temperature data loggers was deployed for two years to compliment longer existing ground temperature time series. Both study sites are located in areas of non-continuous permafrost which contributes to the complexity of the landscape within one scatterometer footprint.
We compared the data sets from the three different sensors for all available time periods to analyze differences in backscatter values and their correlations with each other as well as with the in situ temperature data. Preliminary results show differing correlations throughout the individual scenes as well as within the time series. It can be hypothesized that variations are caused by the different surface characteristics of the landscape types and their freeze/thaw responses as well as different sensor characteristics such as temporal and spatial resolution. Results from this study will advance the understanding of spatial variability within scatterometer footprints and their influence of this data’s applicability to freeze/thaw retrieval in complex terrain.

Area D - Board 473

Numerous evidences show that permafrost is degrading rapidly under current climate conditions, which will also affect the hydrological cycle, ecosystem functioning or built infrastructure. This work attempts to explore the possibility to simulate the permafrost degradation with readily available mapped and remote sensing data, by developing a conceptual model using a geostatistical method based on logistic regression to identify a series of decisive indicators that characterize permafrost decline. The study is performed in the sub-arctic Sheldrake River catchment, located in the discontinuous permafrost region of Nunavik, Northern Quebec. The east-west oriented river drains Sheldrake Lake towards Hudson Bay. Degradation of permafrost due to increasing temperatures is observed in the region since the 1990s.
A conceptual model, based on several indicators, is built to simulate the degradation of permafrost. The process “modelling” can be conceptually seen as a function that maps various inputs (indicators) into an output (permafrost degradation), applying a logistic regression modelling algorithm.
In total, 9 indicators are determined with respect to five important perspectives: the climatic, geologic, hydrologic, topographic and biologic factors. In addition to this functional consideration, the general availability of data is also taken into account as a key criterion for model development. The 9 indicators are altitude, slope, aspect, the distance to the coast, the thickness of permafrost, soil deposit and three vegetation land cover types (lichen, shrub and spruce), which represent density levels of surface coverage. Model output, permafrost degradation since 1957, is categorized in five percentage levels, ranging from less than 5% to more than 50% (Jolivel and Allard, 2013).
Considering the complexity of the natural process of permafrost changes and the variety of the influencing factors, a machine learning approach was used in this work. Logistic regression is used as the classification method to handle categorical and continuous data.
A series of data from different sources are collected and processed, these include a 12m resolution DEM, 5m resolution satellite images from RapidEye, two previous maps of soil deposit and permafrost thickness, the degradation map, and photos with GPS-information taken during substantial fieldwork. A land cover map was created using an object-based classification method of Landsat and RapidEye imagery. The data are processed using a sector-wise and a pixel-wise representation. The sector-wise representation determines 158 representative sectors within the catchment as data points. And the pixel-wise representation considers each single pixel within the catchment as a data point for the modelling. For each data point, there is always a pair of input and output, with nine indicator values as input and a scalar value of the degradation as output. Despite their procedural differences during the modelling process, the sector-wise and pixel-wise representations of data lead to similar results.
First trials of model application already revealed relatively high model accuracy around 66%, with reasonable distributions of predicted degradation degrees (see Fig. 1a). Further refinements of the model are implemented and tested, optimizing the use of available data. The degradation degree for areas with sand deposits (light blue part in Fig. 1a) could finally not be classified well. Considering the difficulty and possible uncertainty of confirming the permafrost presence in the sandy area in the practice, these areas were excluded from further analysis. After elimination of the sandy area from the computation, the accuracy shows a large improvement and finally reached 74.30%.
The influence and the importance of each indicator on the modelling process are analyzed through a backwards greedy feature selection method. Overall, “altitude” and “distance to the coast” are the two most significant indicators, building the main frame of the distribution. The thickness of permafrost also plays a very important role and refines the results of a few critical areas.
In conclusion, a logistic regression based modelling method to estimate the permafrost degradation level is proposed. A high accuracy of 74.3% is reached and all determined indicators proofed as useful for the geostatistical modelling. The study shows a successful and practical way to monitor and simulate permafrost decline using readily available data.

[Figure 1. a) Degradation of permafrost according to a map by Jolivel and Allard (2013), area with sandy deposit are shown in light blue; b) Degradation of permafrost according to a map by Jolivel and Allard (2013); c) Predicted degradation of permafrost; comparison between original and predicted map showed in blue circles]

Area D - Board 474

This paper considers the question of determining frozen/thawed state of 5-cm topsoil layer for seven ground stations in Alaska with latitudes from 65^0 N to 70^0 N (Prudhoe Bay, Imnaviat Creek, Atigun Pass, Coldfoot, Gobblers Knob, Eagle Summit and Upper Nome Creek) by using radar Sentinel 1 C-band data for the period 2016-2017. Determining the status of frozen/thawed soil was carried out in two ways: 1) by using only radar data with finding the backscattering coefficient threshold sigma^0_threshold, when the temperature in the 5-cm topsoil layer falls below 0^0 C, and 2) by using both radar data and ISMN (International Soil Moisture network) ground-based measurements of soil temperature. In the latter case, the correlation between the backscattering coefficient sigma^0 and the soil temperature at different depths from 5 cm to 51 cm is determined. It is shown that the Spearman correlation coefficient is the highest between sigma^0 and the soil temperature at a depth of 5 cm. In the case of high and moderate correlation, regression models are constructed for the relationship between sigma^0 and soil temperature at a depth of 5 cm for the stations Prudhoe Bay, Imnaviat Creek, Upper Nome Creek and Coldfoot with determination coefficients in the case of linear regression, respectively, 0.4, 0.64, 0.62 and 0.46. The found threshold values sigma^0_threshold for polarization VV allowed to construct local maps of frozen and thawed soils. The comparison of the backscatter coefficient threshold values obtained in two ways showed their closeness, which tends to decrease with decreasing correlation between radar and ground ISMN data. Linear regression with R2=0.97 is obtained for the connection sigma^0_threshold VV polarization with place latitude.

Area D - Board 475

The ice-free areas of the South Shetland Islands are dominated by periglacial and glacial processes and landforms, closely associated with deposits of fluvial and coastal origin that are characteristic within the northern Antarctic Peninsula region. These areas are influenced by cold maritime climatic conditions and have experienced the greatest warming within the southern hemisphere over the past sixty years. As a result, freeze-thaw cycles are favoured implying dynamic processes in the active layer of the permafrost, which lead to the formation of different types of surface features. Studies have shown that these ice-free areas have often contained fragile ecosystems with a high biodiversity where many different plant and animal species are concentrated and an active hydrologic cycle during the Austral summer period. It is therefore important to monitor present-day active periglacial and glacial processes linked to recent global changes. Furthermore, determining the spatial distribution of diverse periglacial landforms will be key to detect the presence of permafrost. The objective of this work is to determine and monitor the spatial distribution of landforms of different origin, with special focus on those connected with periglacial processes and the presence of permafrost within ice-free areas. The studied area is located along the southern coast of Livingston Island between Hannah Point and Bernard Point using satellite-borne single and fully polarimetric Synthetic Aperture Radar (SAR) data and techniques.

A fully polarimetric fine RADARSAT-2 image as Single Look Complex data was acquired through the Canadian Space Agency within the framework of the Science and Operational Applications Research Program (Project SOAR-5169) on the 29th of March 2014. Single polarimetric SENTINEL-1 images of 2017 and 2018 were obtained through the European Space Agency. Pre-processing of the SAR data includes radiometric correction, using a speckle noise filter, developed specifically within the team for both SAR single channel images and fully polarimetric images. A digital terrain model is used to carry out the terrain and geometric corrections. Polarimetric parameters are extracted from the RADARSAT-2 data and related to the different surface covers according to the backscattering characteristics received by the sensor within selected field sites. Furthermore, backscattering characteristics are also determined for the single polarimetric SENTINEL-1 data. Several field campaigns carried out during 2013, 2017 and 2018 provided reference for determining the different surface covers. A supervised classification is carried out using Random Forest (RF), a decision tree-based classifier using the different SAR data as well as DEM information. The detailed field observations are further used as reference for the classifier where the importance of each polarimetric parameter is determined in characterizing a particular surface cover. A final step is the validation of the mapping results obtained for the different ice-free areas using both field data and existing maps that have been compiled by the authors team.

Initial results show that periglacial and glacial landforms such as patterned ground, stone fields, block glaciers, ice fronts, moraines as well as raised beach systems and abundant rock outcrops are well identified with the polarimetric RADARSAT-2 and SENTINEL-1 data. The elevation data is also of importance, as this nicely differentiates the areas above certain altitudes in which permafrost exist. Using RF, a variable ranking of the polarimetric parameters for individual terrain classes can be obtained. This is useful when trying to distinguish different classes that bear resemblance to each other such as patterned ground and stone fields. The final mapping is therefore based using the most relevant variables obtained from the different data. The distribution of surface covers obtained with this approach can then be used in time series analysis for future monitoring of changes in ice-free areas of Livingston Island. This would be a novel approach to monitor the active periglacial and glacial processes.

Area D - Board 476

Icings, or aufeis, are common features of permafrost regions. Their presence indicates significant groundwater discharge through fractures and fault zones. Upon freezing, groundwater is retracted from baseflow and stored over winter to feed the streams in summer months. This seasonal runoff redistribution is a key feature of permafrost hydrology.
Sentinel-1 SAR data has emerged as a useful tool in monitoring the areal changes in icings at a weekly timescale. Treatment chain for these data is built using ESA SNAP v.6.0. A final product, suitable for icing detection, is a Dual Pol Difference RGB composite based on sigma0 (in dB) bands in both VV and VH polarization. Icing fields are detectable on the final treated imagery in various shades of blue; darker blue shades are solid and smooth ice, and blueish colors are water-saturated ice or water running over the ice.
The Samokit River valley in Southern Yakutia was selected as a key site for test calculations. This river serves as one of the major water supplies for the Neryungri city, an industrial and mining center of the region, and icing meltwater is supposed to play an important role in total runoff. First, the developed SAR-based model was compared to available optical imagery (Landsat 8) for May and June 2017, when the the icings thermal decay and melting are observed. Areal differences for the five icings observed in this valley vary within 10% range, with SAR consistently underestimating icing areas. Second, the areal change throughout this period was calculated. Three icings out of five had a uniform decay rate throughout the period, while the other two had rapidly melt in May and had slower decay rates in the warmer month of June. Third, the average icing runoff between the dates was evaluated from areal changes. An empirical Sokolov's power law formulation with coefficients from Yoshikawa et al. (2007) was used in this study. An average icing meltwater discharge for the period is 1.1 m3 s-1, and the highest value is 9.91 m3 s-1, falling between 9 May and 14 May, 2017. Concluding, the Sentinal-1 SAR imagery proved to be capable of providing reliable data on icing areas and icing-generated runoff, which can be used in hydrological studies, water supply-related applications, and in monitoring of aufeis flooding events.

Reference:
Yoshikawa K., Hinzman L.D., Kane D.L. (2007) Springs and aufeis (icing) hydrology in Brooks Range, Alaska, J. Geophys. Res. 112, G04S43, doi:10.1029/2006JG000294.

Area D - Board 477

Palsas are a geomorphologic landform consisting of raised beds of peat with an interior permafrost ice lens. They form primarily in flat wetland areas and can be sensitive to increased temperature and precipitation, both of which are anticipated consequences of climate change. A palsa mire complex at the border between Sweden and Finland in the area of Saarikoski is used in this study. In particular, one palsa which has a mound shape, is 4 m in height, and ca 6000 m2 in area is studied. A combination of high and very-high spatial resolution 3D digital surface models (DSMs) as well as spectral data from Sentinel-2 images are used in a multi-scale analysis. Photogrammetric point clouds are used to create 3D data to measure the change in height and volume of the palsa over time. The 3D data are produced from the Swedish national aerial photography program (photographs have a pixel size of 0.5m) and from unmanned aerial vehicle (UAV) images with a resulting pixel size of less than 1 cm. Due to rapid climatic changes, the palsa has undergone some collapse during the short time span of just a few years (2016-2018); this allows recent data acquisitions to be used to observe changes in the palsa. A loss of areal extent (approximately 4%) is seen from analysis of the 2D data, while the 3D data show an apparent loss of volume. The 3D data over the palsa are compared to an equivalent time series of Sentinel-2 data to interpret the spectral changes observed. This multi-scale analysis may offer complementary information when understanding the dynamics observed over similar environments that occur throughout the arctic and sub-arctic. The results show that repeated UAV measurements producing digital surface models can provide a very useful monitoring data source for the changing conditions in palsa mires.

Area D - Board 479

Thematic Open Access data portals foster and support an open data culture in order to reduce knowledge gaps and data uncertainty. We here present the Arctic Permafrost Geospatial Center (APGC), which provides open access, high quality geospatial data in the field of permafrost research. APGC allows easy access to a wide range of permafrost data products including remote sensing, field and modelling data and thus supports multi-scale and interdisciplinary analysis.
The APGC mission is (i) to provide data that is of high usability, significance and impact, and (ii) to facilitate data discovery, data view and (iii) supports metadata documentation and exchange via the APGC data catalogue at apgc.awi.de.
The catalogue structure can host a variety of data models of varying themes, formats, spatial and temporal extents. Data can be searched by location – using spatial keywords or by interactively selecting locations on a base map. Data can further be searched by product type, project, tags, keywords, license type, or data format. Collections of datasets allow users to easily get an overview of the spatial distributions of datasets or their availability in different formats. A WebGIS application is available for most of these data sets, which allows users to explore the data interactively (maps.awi.de).
The Data Catalogue is based on the open source CKAN data catalogue architecture, which uses the metadata standard DCAT. Data is documented according to the fair data principles. Each catalogue entry has a data abstract, data preview and extensive metadata that can be accessed in RDF/XML-, JSON- or Turtle-format. Data can be downloaded directly via link to the publishing data repository such as PANGAEA.
APGC, initially supported by the ERC PETA-CARB and the ESA GlobPermafrost projects currently features over 150 selected datasets mainly from these projects. Data products provide information about surface and subsurface permafrost characteristics in the Arctic, Antarctica, or mountain permafrost areas, e.g., soil temperatures, soil carbon, ground ice, land cover, vegetation, periglacial landforms, subsidence and more. Over 90% of the products are derived from earth observation data including sensors such as Landsat, MODIS, TerraSAR-X, Sentinel and ENVISAT. Users can specifically search for sensor type or spatial resolution. The catalogue also features in-situ measurements and modelling products. Depending on the type of product, data may be downloaded in vector or raster format.
APGC invites submissions from both individual users as well as project consortiums. New data submissions to the catalogue are evaluated according to the following access criteria: permafrost focus, significance and impact, access, quality, and metadata. Data must be archived in a long-term repository to provide easy user access and guarantee long-term availability.

Area D - Board 480

GIS server and desktop GIS technologies support scientific work at all levels, from data collection and data processing to data management and data visualisation. Here we present how the development and publication of scalable WebGIS data services supports the ESA DUE Globpermafrost (2016-2018), and the ESA CCI+ Permafrost (2018-2021) projects, specifically in the interaction with the permafrost community.
Within ESA DUE programs, user feedback is essential to improve the remote sensing products. This is why ESA GlobPermafrost had to focus on methods and infrastructure for data presentation, and established PerSYS (Permafrost Information System). PerSYS became the ESA GlobPermafrost geospatial information service for publishing and visualisation of information and data products to the public. Data products are described and searchable in the PerSYS Data Catalogue, a core component of the Arctic Permafrost Geospatial Centre (APGC), established within the framework of ERC PETA-CARB at AWI. All GlobPermafrost data products will be DOI-registered and archived in the data archive PANGAEA provided by AWI.
The data visualisation employs AWI’s WebGIS-infrastructure maps@awi (http://maps.awi.de), a highly scalable data visualisation unit within AWI’s data workflow framework O2A (from Observation to Archive). GIS services have been created and designed using ArcGIS for Desktop (latest Version) and finally published as a Web Map Service (WMS), an internationally standardized format (Open Geospatial Consortium (OGC)), using ArcGIS for Server. The project-specific data WMS as well as a resolution-specific background map WMS are embedded into a GIS viewer application based on Leaflet, an open-source JavaScript library. The GIS viewer application was adapted to interlink all GlobPermafrost WebGIS projects, and especially to enable their direct accessibility via the GlobPermafrost Overview WebGIS. The PerSys WebGIS is accessible via the GlobPermafrost project webpage and linked to the respective product groups as well as to maps@awi.
WebGIS technology within maps@awi supports the project-specific visualisation of raster and vector data products of diverse spatial resolutions and remote sensing sources. This is a prerequisite for the visualisation of the wide range of GlobPermafrost remote sensing products like: Landsat multispectral index trends (Tasseled Cap Brightness, Greeness, Wetness; Normalized Vegetation Index NDVI), Arctic land cover (e.g., shrub height, vegetation composition), lake ice grounding, InSAR-based land surface deformation, rock glacier velocities and a spatially distributed permafrost model output with permafrost probability and ground temperature per pixel. All WebGIS projects are adapted to the products’ specific spatial scales. For example, the WebGIS ‘Arctic’ visualises the Circum-Artic products. Higher spatial resolution products for rock glacier movements are visualised on regional scales in the WebGIS projects ‘Alps’, ‘Andes’ or ‘Central Asia’. The PerSYS WebGIS also visualises the locations of the WMO GCOS ground monitoring networks of the permafrost community: the Global Terrestrial Network for Permafrost GTN-P managed by the International Permafrost Association IPA. The PerSYS WebGIS has been presented on several User workshops and at conferences, and is being continuously adapted in close interaction with the IPA.

Area D - Board 481

In this study, a tool is created in Google Earth Engine and Earth Engine Apps to monitor area changes of individual lakes in permafrost areas. This is motivated by the close relationship between lakes in these areas and the state of the permafrost beneath, which is an important potential source of carbon emissions.

More specifically, the following questions are addressed:
1. Technical: how can Google Earth Engine and Earth Engine Apps be applied to delineate lakes and their changing areas in permafrost areas? What are the main benefits, challenges and shortcomings of these innovative platforms?
2. Applied: how have Arctic lake areas been changing in recent years and how are they changing now and in the future? The created tool and data can be used to help understand processes related to lake changes and study changes in (sub-) Arctic lake surface area.

Google Earth Engine is a cloud-based platform that makes Google’s computational capacities available to a broad public, including researchers that are not remote sensing-experts. In the light of challenges and opportunities posed by the ever-increasing supply of open access earth observation data, this project serves to explore the possibilities of Google Earth Engine for monitoring lake changes in permafrost areas.

Lake areas within a selected area are first delineated using SAR data from Copernicus Sentinel 1 and optical imagery from Copernicus Sentinel 2 for different times. Based on this, the rate of change is determined for observations of individual lakes. Compared to a pixel-by-pixel comparison, this lake-object based approach can allow additional insight into local hydrological processes and lake area change.

Creating time series of area changes of individual lakes is the type of computationally demanding task for which Google’s cloud infrastructure is expected to be well-suited. We expect to observe limited changes given the small observation period of the Copernicus Sentinels, but hope nevertheless to be able to differentiate between rapid and more gradual changes.

Area D - Board 482

The parameters required by the Global Climate Observing System (GCOS) for the Essential Climate Variable (ECV) Permafrost are Depth of Active Layer (the seasonal thaw layer) and Permafrost Temperature. The ESA CCI+ Permafrost project, funded by the European Space Agency (ESA) 2018-2021, will provide these two ECVs in the permafrost extent simulation, using a set of global satellite data products. Special emphasis for the CCI+ product is placed on satellite-data validation with international and national permafrost monitoring networks and in cooperation with the permafrost community. This approach aims at enhancing confidence in the validity of the resulting new map of potential permafrost extent for the Northern Hemisphere and high-mountain permafrost environments.
The validation and evaluation efforts will also innovatively consider high-mountain permafrost regions, using in-situ observations of ground temperatures, changes in subsurface ice and unfrozen water content, and velocities of permafrost creep, provided by national data-services such as PERMOS in Switzerland. The PERMOS monitoring data will also validate the Earth-Observation derived inventory on rock glacier abundance, extent, and creep, which was developed by the ESA Data User Element (DUE) GlobPermafrost program since 2016 and continues in CCI+ Permafrost. The PERMOS data and the ESA GlobPermafrost rock glacier inventory will support the validation of CCI+ Permafrost products in mountain areas, where the CCI+ Permafrost products contain the highest uncertainties.
To validate the CCI+ Permafrost products, we are collecting and standardizing ground data on permafrost temperature, active layer depth and, additionally, snow depth from regional-scale permafrost programs and research groups, and national monitoring programs such as ROSHYDROMET (Russia). The main validation product will be the database of the Global Terrestrial Network for Permafrost (GTN-P). The World Meteorological Organization (WMO) and GCOS delegated the ground-based monitoring of the ECV Permafrost to GTN-P managed by the International Permafrost Association (IPA). GTN-P/IPA established the Thermal State of Monitoring (TSP) and the Circumpolar Active Layer Monitoring program (CALM), including standards for measurements and data collection.
We are going to present an overview of temporal and geographical coverage of the CCI+ Permafrost validation dataset. This comprehensive dataset includes variable timeframes from hourly over annually to sporadic measurements and covers a wide range of different vegetation and permafrost types. Prior to validation, however, trouble-shooting such as correcting inaccurate coordinates leading to misplaced boreholes, is necessary. Furthermore, we present our steps of data mining, standardization, data optimization and metadata completion.
This optimized and standardized validation data set will be supplied within the CCI+ Climate Research Data Package (CRDP) and will thus be publicly available for validation also for the broader climate science community.

Area D - Board 483

Permafrost is a key element of the terrestrial Cryosphere and has been recognized as an Essential Climate Variable by the Global Climate Observing System. Permafrost state and extent is affected by a warming climate in a variety of ways. While widespread permafrost degradation will be likely to occur in this century, local permafrost degradation due to disturbances such as thermokarst and thermal erosion has been a variable background signal since the end of the last ice age. Remote sensing can help to untangle the role of local disturbances from a potential widespread climatic signal if spatially and temporally continuous, high resolution data can be used to understand landscape changes in permafrost regions.
The European Space Agency launched the GlobPermafrost initiative (2016-2019) to develop, validate and implement information products to support the research communities and related international organisations in their work on understanding permafrost better by integration of Earth Observation data. As part of this project, the Hot Spot Regions of Permafrost Change (HRPC) product was developed in order to maps changes in surface characteristics across four large regional transects using the full Landsat archive (TM, ETM+, OLI) for the peak summer season (July, August) from 1999 to 2014. The 30m-resolution HRPC product provides trends of multi-spectral indices such as Tasselled Cap brightness, greenness, wetness and normalized indices (NDVI, NDWI, NDMI) in four regional transects totaling about 2.5 Mkm2 spanning the panarctic permafrost domain (West Siberia, East Siberia, Alaska, and NE Canada). Changes in the indices can be linked to both gradual (press disturbance) and rapid (pulse disturbances) land surface changes. Press disturbances are for example shrub expansion, wetting, drying, or active layer deepening. Pulse disturbances include thermokarst and thermo-erosion, lake formation, expansion, and shrinkage, erosion, or wildfires. These disturbances all have an impact on the thermal regime of the ground and thus potentially the state of permafrost underneath. Their trends can be a potential indicator for permafrost vulnerability or actual permafrost change.
With the new ESA Permafrost CCI+project, funded by the European Space Agency (ESA) 2018-2021, the transient permafrost model CryoGrid 2 developed at the University of Oslo will be used to assess the panarctic permafrost region ground thermal regime specifically by integrating remotely sensed data sets of Land Surface Temperature (LST), Snow Water Equivalent (SWE) and landcover. The production of consistent permafrost maps and active layer thickness at several epochs at targeted spatial scales between 10 and 1km meets the requirements of the climate modelling community.
In this study we will present first results from cross-analysing the high-resolution HRPC trends of disturbance-driven land cover changes in permafrost regions for selected sites with the results of MODIS-based LST and transient permafrost modeling results from ESA CCI+ Permafrost. Development of this science use case in ESA Permafrost CCI+ will allow better understanding of the role of local sub-grid disturbances for broad-scale permafrost vulnerability and change patterns.

Area D - Board 484

Rock glaciers are common in arid or semi-arid mountain environments and can contain significant amounts of ice, and may represent an important water resource in water-stressed areas in the future. While advances have been made in the creation and maintenance of up-to-date regional to global scale glacier inventories using satellite data, the methods involved (namely multispectral band ratios and SAR coherence) are incapable of distinguishing rock glaciers from the surrounding terrain. Consequently, most rock glacier inventories are based on manual interpretation of very-high resolution satellite or aerial imagery, and as such are often out of date, inconsistent, or incomplete. Advances in machine learning capabilities, as well as the availability of high-resolution optical and radar satellite data allow for more sophisticated classification methods. Here, we present preliminary results of an automated method that utilises deep learning (also known as convolutional neural networks) and SAR interferometry (DInSAR) within an object-based image analysis (OBIA) relying on Sentinel 1 and 2 data along with a Digital Elevation Model. We trial our method in two contrasting periglacial environments, the Poiqu Basin of Tibet, Central Himalayas, and the central Swiss Alps. The former is considerably more arid and in many ways more challenging, while in the Swiss Alps, vegetation is present at the elevation ranges of many rock-glaciers. Deep-learning is capable of utilising subtle textures, patterns, and other features found within images, and outputs a “heatmap” where each pixel ranges from 1 (definitely a rock glacier) to 0 (definitely not a rock glacier). This heatmap contains many false positives however, such as landslide deposits or rock falls. The rock glaciers in the Swiss Alps were easier for the deep learning to identify, as the varying vegetation cover was more noticeable in the spectral values of the Sentinel 2 imagery. We also used Sentinel 1 IW DInSAR displacement data to remove as many erroneous results as possible. DInSAR however struggled due to lack of radar coherence in areas with vegetation or surface water, as such DInSAR results were less reliable in the Central Alps compared to the Poiqu Basin. In both cases, we refined and reshaped the resulting rock glacier outlines within an Object Based Image Analysis (OBIA). Contextual and spatial information of segmentation-derived objects was used to remove notecable false positives due to their shape or compactness. We believe our method shows promise as a transferable and reliable method for rock glacier classification, incorporating the entire procedure within an OBIA allows for a degree of flexibility that we hope will allow the method to be further developed and applied to different periglacial environments.

Area D - Board 485

The behaviour of periglacial landforms, such as rock glaciers, or other permafrost related processes can be of high significance for paleoclimatic reconstructions but also a proxy for ongoing climatic changes. In particular, knowledge on surface displacements rates and patterns is an added value for a better understanding of periglacial environment dynamics. However, studying such displacements and their evolution is challenging due to often difficult environmental conditions and remoteness. EO data is of great help in this respect, and the combination of remote sensing approaches with geological and geomorphological information is nowadays more and more applied in such investigations. Among remote sensing techniques, satellite based Synthetic Aperture Radar Differential Interferometry (DInSAR) can be used to measure ground displacements, potentially allowing for millimetric accuracy with current time series techniques and corrections from atmospheric effects.
Expected displacements of active rock glaciers are gravitationally driven and irreversible, while active permafrost layers predominantly show seasonal reversible variations of uplift and subsidence. Although atmospheric effects are of lower amplitude than the large linear trends associated with rock glacier creep, atmospheric corrections are essential for the detection of reversible deformation as well as improved accuracy of the linear trends.
In this work, we study a 8000 km2 area in northwestern Bhutan, with elevations between 1000 and 5000 m a.s.l.. Our analyses combine information obtained from GoogleEarth optical images and from multitemporal time series analyses. Radar interferograms from 6 different datasets (two Envisat tracks and 4 ALOS-1 frames) were analysed individually for visual interpretations, as well as exploited for the generation of displacement time series based on a small baseline approach (SBAS).
We reduced the effect of the atmosphere and refer all cumulative displacements maps to common stable areas by using the GACOS atmospheric correction tool and an by applying an empirical estimation of residual tropospheric delays and long-wavelength ramps. The multi-annual trends over rock glaciers and the amplitudes and the timings of the seasonal displacements over other permafrost areas were then analysed with respect to elevations, aspect, mean annual temperatures and morphological expressions.
167 rock glaciers were identified through available optical images. Of these, 52% and 72% have been classified as active through the analysis of single interferograms and SBAS results respectively. Only 2.4% of the rock glaciers mapped in optical images and 15% of those mapped in the interferograms are not covered by information obtained through SBAS. Our analyses indicate that rock glaciers in the region during our observation period (between February 2007 and February 2011) creep with estimated downslope displacement rates ranging from a few mm/year to 70 cm/year. These velocities are notably lower than the m/years of displacement rates observed for some rock glaciers in other mountain belts (e.g. the Alps) and are likely related to mean annual temperature conditions.
Moreover, we were able to single out areas of seasonal reversible deformation lying on low roughness, fine sediment covered valley floors, with gentle slope angles, generally less than 15°. Their spatial extent ranges between 400 m and 3 km in radius and their elevations vary between 4100 and 5100 m a.s.l.. The amplitudes of their seasonal cycles are as high as 15 mm and show a predominant vertical component, with uplift in winter and subsidence in summer, which is compatible with freeze-thaw cycles. These areas show almost no long-term linear trends or increase in the seasonal amplitude, which may indicate that, in the period of observation, no long-term change of permafrost conditions is detected in the region.

Area D - Board 486

Relevance of the study area:

The Tibetan Plateau (TP) is also referred to as the Third Pole, due to being the third largest storage of frozen freshwater on the planet. It is the source of many of the world’s largest river systems and supplies freshwater to over 1.6 billion people (Cuo & Zhang, 2017). As such it is currently being studied extensively, especially in regards to the effects of climate change and global warming on this vital system. The TP’s rising air temperature of 0.3 °C per decade (Yao 2007) has led to shrinking glaciers and permafrost degradation. The Nam Co, one of the largest endorheic lakes on the TP, has grown in size by almost 5 % over the last 40 years (Mao et al., 2018), while glaciers in the area are retreating simultaneously. Due to its accessibility and significant reaction to climate change, the Nam Co catchment was selected as the primary study area for the TransTiP research group (https://www.tu-braunschweig.de/irtg-transtip), a newly formed international graduate school focused on studying the effects of climate change on the TP.


Using InSAR time-series analysis to detect surface changes:

Glacial retreat is relatively easy to track via optical satellite data. However, the recent degradation of permafrost is often much harder to investigate on a large scale. Interferometric Synthetic Aperture Radar (InSAR) time-series data is better suited to study those periglacial processes, which often cause annual deformation on a millimeter to decimeter scale. The lack of vegetation and the limited amount of snow cover in the study area reduces the risk of decorrelation. This allows time-series analysis of InSAR data with very few temporal gaps and good spatial coverage. The atmospheric system at Nam Co poses some challenges, as strong monsoonal rains may cause decorrelation during the summer months. In this study we are using the Small BAseline Subset (SBAS) technique to track periglacial landform changes within the Qugaqie Basin at Nam Co. We employ ESA’s Sentinel-1 Radar satellite and validation will be performed with TerraSAR-X ScanSAR data. A geomorphological map of periglacial landforms, like rockglaciers and protalus ramparts, is used to corroborate the surface changes displayed in the time-series data.


Preliminary results:

The high resolution time-series data shows mean surface movement rates of 10 – 30 mm/yr in the Qugaqie basin. Some slopes creep with mean rates of up to 50 mm/yr. Areas with fast movement rates, which are marked as ‘Potential Ground Ice’ in Figure 1, match the mapped periglacial landforms of the Qugaqie basin. The annual freezing and thawing cycle of the active layer is also clearly represented within the time-series data. The exact onset and duration of both freezing and thawing periods, as well as the amplitude of the resulting surface displacement vary throughout the basin. Most areas in the bottom of the main valley experience uplift through freezing of the active layer from September to January and subsidence through subsequent thawing from April to August. The amplitude of this annual cycle is up to 7 mm. This fits the monsoonal climate of this area, as the thawing period corresponds to both the rise in temperature and precipitation and the freezing period begins once temperature and precipitation start to drop again.


Outlook:

The next step in this research will be to validate the Sentinel-1 time-series results. This will be done in two ways: Firstly by processing TerraSAR-X ScanSAR data of the same area and comparing the results of both datasets. To further validate the linear trend of our satellite data, terrestrial laser scans of a few key areas were recorded in summer 2018, which will be repeated in September 2019. This will provide an additional dataset for validation, which is not influenced by atmospheric effects. Additionally, this dataset partially closes the gap of missing surface deformation rates in the northern direction, which are difficult to ascertain from satellite data alone.


References:

Cuo, L., & Zhang, Y. (2017). Spatial patterns of wet season precipitation vertical gradients on the Tibetan Plateau and the surroundings. Scientific Reports, 7(1), 5057.

Mao, D., Wang, Z., Yang, H., Li, H., Thompson, J. R., Li, L., ... & Wu, J. (2018). Impacts of climate change on Tibetan lakes: Patterns and processes. Remote Sensing, 10(3), 358.

Yao, T., Pu, J., Lu, A., Wang, Y., & Yu, W. (2007). Recent glacial retreat and its impact on hydrological processes on the Tibetan Plateau, China, and surrounding regions. Arctic, Antarctic, and Alpine Research, 39(4), 642-650.


Figure:

Figure 1: Linear surface velocity in Qugaqie basin in east-west (A) and vertical (B) directions from 2015 – August 2018 (© Copernicus Sentinel Data [2015-2018]) above a shaded TanDEM-X DEM (© DLR [2016]). Areas with significant surface motion, which could be caused by ground ice, are outlined in black.

Area D - Board 487