Cryosphere: Operational Applications & Services

Climate warming in recent decades has put a significant part of the Arctic permafrost at risk of extensive thawing, which is threatening the existing infrastructure such as roads, rail beds, airport runways, pipelines, and buildings. The heat balance of underlying permafrost is gradually shifting towards greater instability. Mitigation measures are needed to avoid permafrost thawing and degradation underneath and around important infrastructure to prevent gradual infrastructure deformation and destruction. Thus, there is a growing demand for monitoring the stability of existing and the impact on newly planned infrastructure.

The synergistic exploitation potential of multi-frequency SAR data (C- and X- band) is studied at three Canadian test sites, located in different permafrost zones, in 2014 and 2015. It is assumed that they show different thawing and surface deformation processes. They are monitored with the TerraSAR-X and RADARSAT-2 satellites at different incidence angles using their high resolution spotlight modes. The incidence angles varied between 24°and 48°, the polarisation is HH:

Radarsat-2 Spotlight A (SLA) mode at 1.3m x 0.4m

TerraSAR-X High Resolution SpotLight (HS) at 0.6m x 1.1m, and

TerraSAR-X Staring SpotLight (ST) mode at 0.6m x 0.24m.

The high temporal coverage of the SAR acquisitions helped to analyze the complex dynamics in the permafrost and to differentiate between temporal deformations due to frequent freeze/thaw cycles of the active layer and long-term deformation due to permafrost degradation as a result of global warming. Interferometric analysis (InSAR) using Small Baseline (SBAS) techniques, shows the following:

Temporal de-correlations of the SAR signal due to snow cover, change of snow cover and snow wetness (dielectric properties) occur in winter and early spring. Particularly quick changes happen in springtime due to the onset of thawing and interlinked gradual changes of land cover / vegetation.

InSAR SBAS analysis is feasible for the majority of the infrastructure items. Objects with a strong temporal de-correlation, such as unpaved roads or airport landing strips, cannot be measured.

X- or C-band show very similar linear displacement measurements. However, X-band provides a higher degree of spatial detail and a higher temporal frequency, while C-band provides a larger spatial coverage with more coherent image points. SAR data acquired at shallow incidence angles provided less coherent information.

The remaining area is mapped by land cover classification. Both TerraSAR-X and Radarsat-2 data are employed separately to compare the suitability of the available acquisition dates, orbit directions, incidence angles and spatial resolution modes. For each satellite, sets of four raster images per image pair are analyzed by multivariate statistical methods: 1) the radiometrically calibrated and orthorectified amplitudes, 2) the phase coherence between image pairs, 3) the image ratios between consecutive acquisitions and the 4) coefficient of variation for each image (3x3 window) which led to the following intermediate results:

Spring and summer coherence images show best class separability in both TerraSAR-X and Radarsat-2.

Both C- and X-band are suitable to map 10-12 similar land cover classes. Spatial coverage and resolution differ and the imaging modes shall be employed according to the desired map scale or minimum mapping unit.

The spatial degree of detail of TerraSAR-X Staring Spotlight is especially high among man made features.

The resulting land cover and InSAR based surface deformation maps are interpreted jointly with a DEM, surficial geology, hydrological regimes etc., to create surface deformation risk maps. This information will help to plan compensation measures and to secure new infrastructures in permafrost regions.

There is a long history of the use of satellite SAR for ship detection and surveillance. Satellite SAR’s large swath and ease of data collection regardless of geographic location makes it particularly attractive for maritime surveillance. SAR is also used for iceberg surveillance to support mapping of maritime hazards. As Arctic sea ice continues to decline, there is growing commercial interest in northern transportation and resource exploration and development. C-CORE has been researching the potential of SAR to monitor iceberg since 1997 and has exploited the capability by offering SAR-based surveillance services to various clients, primarily the oil and gas (O&G) industry. This capability is complementary to ship surveillance in ice frequented waters, since eliminating false positives due to icebergs is generally a requirement. In the context of Defence, Canada’s Department of National Defence’s (DND’s) presently uses RADARSAT-2 data for its maritime surveillance program (Polar Epsilon) and data from RADARSAT Constellation Mission (RCM) will be used when that mission is launched in 2018. Thus, there is synergy between C-CORE’s O&G space-based iceberg surveillance program and Polar Epsilon. To this end, the authors are collaborating on a joint O&G and Defence research project to improve SAR-based ship and iceberg surveillance.

SAR-based surveillance in ice frequented waters tends to focus on optimizing detection and the ability to discriminate between different types of targets. The authors have already demonstrated the ability for dual-polarization (DP) and quad polarization (QP) SAR to improve detection and discrimination with data from ENVISAT, RADARSAT-2 and TerraSAR-X. While QP SAR provides the best detection and discrimination relative to other modes, its relatively narrow swath (25 to 50 km) on RADARSAT-2 limits its utility for surveillance applications. Compact polarimetry (CP), a compromise to QP, is a dual polarization mode where circularly polarized waves are transmitted and horizontal and vertical polarized waves are received. The information content of this configuration is greater than standard DP, but less than QP. Both CP and DP are available on most modes of RCM, so there is significant interest in the value of CP for ship and iceberg monitoring, relative to its performance with standard DP.

There are many possible approaches to the analysis of CP image data, with the dominant approach focusing on the use of parameters derived from the CP Stokes vector. Another approach is to use the CP data to estimate components of the quad-pol covariance matrix and use these reconstructed PQ data in existing linear polarization algorithms. The authors are investigating these approaches with RCM data generated with a simulator provided by the Canada Centre for Remote Sensing; these data have been derived from ground validated RADARSAT-2 data of ship and icebergs, modified in resolution, number of looks and noise equivalent sigma zero to match various RCM modes.

Receiver operating characteristics (ROC) are being derived for various RCM CP and DP modes using iceberg and ship targets. The ROC curves illustrate relative performance of the various modes for detection, thus allowing an analysis of the best SAR modes for optimizing probability of detection. In addition, various ship/iceberg classifiers are being derived, including a one class classifier to discriminate a target as a ship or an iceberg (depending on the application) or a two class discriminant, to discriminate ship and icebergs from one another. The one class classifier will use a clustering approach (in feature space) or support vector machine (clustering in projected space), while the two-class discrimination will use a quadratic discriminant classifier, a support vector machine or an ensemble classifier. The paper and presentation will include these results and discuss the utility of each RCM mode for these maritime applications. In particular, the paper will illustrate the value in using CP over standard DP.

This paper investigates the potential improvement of iceberg detection and characterization using radar polarimetry. The interest in detection and monitoring of icebergs has a number of reasons. The most obvious challenge is the fact that they pose a danger to ships and offshore structures. While there is a great variety in iceberg shapes, sizes, and overall geometries, all icebergs, especially small ones, i.e., growlers, can present potential hazards to ships because they are harder to spot [1,2]. As one of the most promising areas for future oil development, much attention has started to turn to the Arctic. In the high Arctic sea, there are a number of icebergs floating or grounded in sea and/or sea ice, which may cause tremendous damage to the undersea or subsea pipelines and production facilities [3].

A number of various satellite sensors have been used for monitoring icebergs. The use of data from optical sensors requires suitable cloud and light conditions. This restriction does not hold for synthetic aperture radar (SAR) imagery and hence iceberg detection generally is an important application of SAR in polar regions. In particular, full polarimetry offers better detection performance over single polarization [4].

In April and September 2015, UiT - The Arctic University of Norway and Northern research institute (Norut) conducted a combined satellite and arial photography campaign on Kongsfjorden in Svalbard.  The aim of this campaign was to collect RADARSAT-2 fine-quad scenes SAR data and near-coincident remotely piloted aircraft systems (RPAS)/unmanned aerial vehicle (UAV) photos. RADARSAT-2 C-band data containing icebergs and glowers in ocean and/or sea ice background are evaluated with corresponding auxiliary data. This auxiliary data and RPAS imagery will be used to visually identify any actual icebergs and growlers and be used as ground truth in the evaluation study described below. The second area of interest is over Hopen, an island in the southeastern part of Svalbard archipelago. The Norwegian Meteorological Institute’s manned weather station at Hopen has observed a number of icebergs and growlers. Radarsat-2 fine-quad scenes are used for iceberg detection and characterization in this area.

To detect icebergs within RADARSAT-2 imagery, the backscatter caused by icebergs must be distinguishable from the ocean clutter. There are several factors that affect the intensity of ocean clutter in SAR imagery, most notably general sea-state, radar look direction, and polarization state [3]. To determine target detection performance, the effects that these variables have on ocean clutter intensity are considered.

The appearance of icebergs is evaluated with respect to scattering mechanisms and polarimetric information [4]. To be able to detect a target in a SAR image, a significant contrast between marine target and background is needed. In this study, peak-to-background ratios (PBR) and target-to-background ratios (TBR) are used to evaluate the contrasts for different target types and background classes for different polarization channels and target decomposition components, and possible relations between contrast measures and incidence angle, iceberg shape and meteorological conditions are investigated. This study includes a refinement of detection capabilities of data with heterogeneous background clutter (sea ice), and investigations of sub-aperture processing [5].

REFERENCES

[1] M. Denbina, Iceberg Detection Using Compact Polarimetric Synthetic Aperture Radar, University of Calgary, PhD dissertation, University of Calgary, Canada, Aug. 2014

[2]  C. Wesche and W. Dierking, “Iceberg signatures and detection in SAR images in two test regions of the Weddell Sea, Antarctica,” J. Glaciol., vol. 58, no. 208, pp. 325–339, Apr. 2012.

[3] D. Power, J. Youden, K. Lane, C. Randell, D. Flett, “Iceberg Detection Capabilities of RADARSAT Synthetic Aperature Radar,” Can. J. Remote Sensing, 27 (5), pp. 476–486, 2001.

 [4] W. Dierking and C. Wesche, “C-Band radar polarimetry—useful for detection of icebergs in sea ice?,” IEEE Transactionson on Geosci. Remote Sens, 52(1):25-37, Jan. 2014.

[5] C. Brekke, C., S.N. Anfinsen and Y. Larsen:  “Subband extraction strategies in ship detection with the subaperture cross-correlation magnitude,” IEEE Geosci. Remote Sens. Lett., pp. 786-790, Apr. 2013.

Sea ice monitoring is an important field of scientific research and relevant to marine operational applications. Presence of sea ice causes a serious problem for ship traffic, especially for non-ice-strengthened vessels. In areas with high ice concentration of thick ice and hazardous ice features (e.g., deformed ice and icebergs) even the icebreakers will have difficulties in their operations. Therefore the information about sea ice and iceberg conditions is required for safe ship navigation. Satellite remote sensing imagery is useful for monitoring sea ice, identifying and tracking ice features over broad spatial scales. At the same time, current satellites have limited ability to provide important sea ice characteristics such as thickness and deformation with required temporal frequency, spatial scale and coverage. The present work investigates the retrieval of sea ice parameters relevant to ship navigation from multisource data including satellite data and sea ice model outputs. The Los Alamos sea ice model (CICE) was implemented on a regional scale with the spatial resolution up to 2km. In addition to the high resolution, the advantage of the model is the possibility of including oceanographic (e.g. currents) and climatological parameters to determine the dynamic and thermodynamic behaviour of sea ice. Assimilation of satellite data in the model improves mapping accuracy of sea ice parameters.

A simulation of first year ice was performed over the Baffin Bay region and the Labrador Sea. Important parameters, such as, ice thickness, ice edge, ridging area and concentration were extracted using model and satellite data. Comparison of modeling results with ice parameters extracted from the satellite measurements (microwave radiometer, SMOS and Cryosat-2) demonstrated a good agreement for ice concentration and thickness.

In the Arctic Ocean hazardous ice features, such as ridges, multi-year hummock fields and glacier ice (icebergs and ice islands) within multiyear ice were analysed using satellite data. Techniques for detecting and monitoring hazardous ice features from optical and synthetic aperture radar (SAR) imagery were developed. The method developed to study ice deformation features involved 3D modeling and visualization of spatial characteristics and was validated with detailed information from very high resolution (0.5m) optical satellites. The problem of detecting and discriminating icebergs and ships with the presence of sea ice was investigated using polarimetric RADARSAT-2 data. Several data classifiers were applied to improve accuracy of automated ship-iceberg discrimination based on geometrical and polarimetric decomposition parameters. The results were validated with optical satellite and auxiliary data. The highest performance (more than 80% of classification rate) was achieved with using support vector machine technique.

Sea ice drift information with an accuracy that allows also ice deformation (divergence, shear, vorticity) to be derived is being operationally generated in the Copernicus Marine Environment Monitoring Service (CMEMS).

The method is based on 2-dimensional digital cross correlation where subsections of 2 consecutive images (typically 12-36h apart) are compared and the ice displacement defined as the shift in location of images 1 that maximizes the cross correlation with image 2. The method is also known as Maximum Cross Correlation or MCC.

Implementation was carried out in the context of the PolarView project in 2007 when large volumes of ENVISAT ASAR images of the Polar regions became available during the International Polar Year.  A dataset of daily ice drift vectors of the Polar Regions (North and South) is now available covering the time period from 2007 to the present time.

In 2009 the processing became part of the GMES Marine Core Service MyOcean and when ENVISAT seized operation in 2012, this enabled a switch to daily RADARSAT-2 coverage of key regions in both hemispheres covered by the GMES Space Component Data Access (GSC-DA) grant.

From October 2014, the data provision has switched to Sentinel-1 the source for SAR data, and the daily coverage of ice drift is now similar to what was reached with ENVISAT during the IPY.  With the Launch of Sentinel-1B in 2016, daily coverage of most of the Arctic Ocean will become possible.

Already today approximately 10.000 Sentinel-1A image pairs are matched every month in the processing system.

The quality of the ice drift vectors are routinely verified against GPS locations of drift buoys and the RMS difference between the baseline product available through the Copernicus Marine Environment Monitoring Service data portal and GPS drifters is ~500 meters per day. A significant part of this RMS difference can be ascribed to the different nature of a point measurement and an area measurement.

This accuracy is sufficient to support the generation of daily maps of ice divergence, shear and vorticity as the spatial derivatives of the ice drift field.

The deformation fields are produced in the FP7 POLAR ICE project which develops methods for downstream distribution of ice related information to end-users in Polar Regions.

The presentation will provide more details on the processing system and examples of the products and the POLAR ICE downstream service.