Final Report Summary - SIDARUS (Sea Ice Downstream Services for Arctic and Antarctic Users and Stakeholders)
Executive Summary:
The objective of SIDARUS has been to develop new sea ice data products from satellite, airborne and underwater platforms as well as ice forecasting from models in order to improve sea ice monitoring services. In addition to analysis of satellite and airborne data, the project has analyzed in situ, airborne and under-ice data from previous and new field campaigns. These are essential data for validation of satellite retrievals of sea ice parameters before the satellite products can be implemented as services for users. The results of SIDARUS are obtained in the following six categories: (1) SAR based products on regional and local scale, where new algorithms for ice classification, ice edge detection, ice drift, ice deformation and iceberg detection have been developed. The SAR data have been obtained from ENVISAT, Radarsat-2 and TerraSAR-X under agreement with GMES. The main application of these products is to improve the operational ice monitoring services; (2) A sea ice albedo retrieval algorithm based on optical satellite data, mainly MERIS data form ENVISAT, has been developed. The algorithm has been implemented to generate maps of sea ice albedo and meltpond fraction in the period from spring to autumn. The main application of these products will be in atmosphere-ice-ocean modelling. The main users group is therefore sea ice and climate modellers who need improved parametrization of albedo and meltpond fraction in their models; (3) Sea ice thickness analysis has included a retrieval algorithm for thin ice (< 0.5 m) based on 1.4 GHz passive microwave data from the recently launched ESA SMOS mission. Measurement of thin ice is a very innovative method, which will be supplementary to the thickness retrievals from altimeter data from IceSat and CryoSat. Altimeter measurement of sea ice freeboard can be used to quantify the thicker part of the ice cover, typically multiyear ice and firstyear ice thicker than about 1 m. In order to validate the ice thickness retrievals, several non-satellite methods have been used. These include profiles of submarine, AUV and mooring-based upward-looking sonar data, electromagnetic induction measurements from helicopters, and in situ data from field experiments. Studies have been conducted to validate thickness retrievals from altimeters, but there is need for considerable more validation work. Ice thickness data are needed by many users, ranging from sea ice and climate modelers to operational ice services, ice navigators, polar shipping and offshore industry working in the Arctic. (4) An ice-ocean forecasting model on regional scale has implemented in the Barents and Kara Sea with improved representation of sea ice dynamics in the marginal ice zone. This model is will support offshore operators and ships that need better ice forecasting in this region. SIDARUS has also supported development of a commercial iceberg service for ships in the Antarctica. The service combines observed icebergs from satellite data with a regional Antarctic circumpolar current model in order to provide forecasts of icebergs. The main customer for this service so far is the Vendee Global race, but other ships operating in Antarctica can also benefit form the service. (5) SIDARUS has demonstrated how ARGOS position data from tagged animals can be combined with regional sea ice maps, where SAR data have been used. These products are important for studying the behaviour of marine mammals in the polar regions. The main user group are environmental monitoring and wildlife protection institutions. (6) Several field campaigns in the Arctic have been supporter by SIDARUS in order to improve collection of validation data for the satellite retrievals. The campaigns have provided data from in situ, underwater and airborne platforms, in particular for ice thickness estimation.
The data sets and products have been demonstrated to users working in climate research, marine safety and environmental monitoring. The results of SIDARUS will be used to extend the present GMES marine services with new satellite-derived sea ice products, ice forecasting from regional models and validation of sea ice products using non-satellite data. The demand for improved sea ice information in the Arctic and Antarctic by many user groups is growing as a result of climate change and its impact on environment and human activities. The presently observed reduction of the Arctic sea ice extent, in particular during the summer months and an increasing demand for natural resources are key mechanisms driving human activities in these areas.
Project Context and Objectives:
Context of SIDARUS
The demand for sea ice information in Arctic and Antarctic seas by many user groups is growing as a result of climate change and its impact on environment and human activities. The presently observed reduction of the Arctic sea ice extent, in particular during the summer months and an increasing demand for natural resources are key mechanisms driving human activities in these areas. The EU has started to develop a policy for the Arctic where protection of the environment and sustainable use of resources are issues of high priority. The expected growth in ship traffic, oil and gas exploration, fisheries and tourism in the coming years will increase the risk for accidents affecting environment, health, safety and economy in this unique and vulnerable region.
Sea ice is a crucial component of the climate system, influenced by conditions in both the atmosphere and ocean. Variations in sea ice may in turn modulate climate by altering the surface albedo; the exchange of heat, moisture, and momentum between the atmosphere and ocean; and the upper ocean stratification in areas of deep-water formation. Recent reductions and changes in the Arctic sea ice cover have already shown severe impact on the environment and climate in the Arctic, with predictions that the changes will accelerate in the next decades. The climate change in the Arctic, and its impact on human activities and environment, has generated increasing demand for sea ice information on past, present and future scenarios. Sea ice data is required by several user groups: marine transportation operators and regulators, meteorological organizations, Search and Rescue organizations, defence and security agencies, resource developers (oil, gas, timber, minerals, fish), policy-makers, marine engineers, populations living and working in ice-affected regions, and environmental institutions working with sea ice habitat protection. Last but not least, climate modellers need significantly improved sea ice representation in the climate models.
Sea ice observation from satellites has been done for more than three decades. Several sensors and retrieval methods have been developed and successfully utilized to measure sea ice area, concentration and drift (e.g. Breivik et al., 2009). Today operational sea ice monitoring and analysis is fully dependent on use of satellite data. However, new and improved satellite systems require further development of sea ice remote sensing. Monitoring of sea ice concentration, extent, and drift using satellite data is provided by the Marine Core services of MyOcean and other services. Use of SAR data for operational sea ice monitoring has been introduced in MyOcean to produce regional ice charts and ice drift maps (http://www.myocean.eu.org). In the GSE project PolarView, several sea ice services are provided, as described at http://www.polarview.org/services/. SIDARUS will provide new sea ice products for climate, marine safety and protection of sea ice habitats as well as in situ data for validation of satellite-derived sea ice products.
Sea ice monitoring and other metocean services (meteorology, oceanography and sea ice) are presently providing large- and regional-scale ice charts and forecasting for the Arctic and Antarctic seas. GMES ice services deliver satellite- and model-based products that aim at general applications for a wide range of users. However, high latitude users such as offshore industry and Arctic-Antarctic shipping require more specific services in addition to the present GMES services. For climate research and prediction, there is a set of essential climate variables (ECVs) where satellites can provide important data. In addition to ice area and ice drift, which are already monitored by existing GMES services, data on sea ice thickness, leads/polynyas, surface albedo, and snow cover are needed, but these variables are not included in the services delivered by MyOcean or other GMES projects. SIDARUS has de eloped new sea ice service products including high-resolution sea ice data to support operators on regional and local scale as well as large scale information for climate monitoring. SIDARUS has developed SAR-based products as preparation for operational use of Sentinel-1 data for sea ice and iceberg monitoring. Furthermore, SIDARUS has provided validation data for satellite-retrieved sea ice products, and implemented regional ice forecasting using high-resolution ice –ocean models in order to meet the requirements from offshore operators and shipping activities in different regions of the Arctic.
Objectives of SIDARUS
The following specific objectives of SIDARUS have been to
(1) Develop sea ice classification and iceberg detection using new high-resolution SAR images with different frequency and polarization, and implement a monitoring service based on SAR data from Sentinel-1
(2) Provide sea ice thickness data for thin ice (<≈ 0.5 m) using the new 1.4 GHz passive microwave data from SMOS, as a complement to the ice thickness data from CryoSat
(3) Collect and analyze data on sea ice thickness and other ice parameters data from airborne, in situ and underwater platform experiments in order to validate satellite retrievals and fill gaps in sea ice observations that satellite data cannot provide
(4) Develop and validate sea ice albedo and meltpond retrieval from multi-spectral optical images, e.g MODIS on the EOS platforms and MERIS and AATSR on ENVISAT
(5) Provide integrated maps of marine mammal tracks from ARGOSS data and sea ice maps from satellite data
(6) Implement a high-resolution operational ice-ocean model in order to provide sea ice and iceberg forecasts on regional and local scale
(7) Demonstrate sea ice monitoring and forecasting services to user groups by integration of observational products from several platforms and simulation/forecasting products from global and regional models using GIS and web technology
(8) Plan sustainable sea ice downstream services for GMES, consisting of both public domain and commercial products
Project Results:
WP1 User interaction
The consortium has been in contact with users of the sea ice products and services throughout the project period. The users include representatives for each of the five application areas where products are being developed: (1) High-resolution sea ice and iceberg mapping by SAR to support marine transportation, offshore operations, and ice-weather forecasting services in polar regions. (2) Sea ice albedo from optical sensors to support sea ice modelling and climate research, (3) Sea ice thickness from satellite data, airborne surveys and other observing platforms providing data for climate research and operational users. (4) Ice forecasting based on numerical models and satellite data for marine safety. (5) ARGOS tracking of marine mammals combined with sea ice maps to support environmental management, wildlife protection. Products in all these five categories have been developed and disseminated to users through workshops and conferences. In addition the consortium members have had meetings with identified users throughout the project period. In some cases users have directly supported the downstream service development through addition projects, supplementing the work in SIDARUS.
WP2 Field campaigns with data collection
Sea ice data from several field campaigns have been obtained in the period. In particular in situ ice thickness data have been collected because these data are important for validation of the satellite retrievals from satellite altimeter data. The ice thickness data used in the project includes airborne electromagnetic measurements of ice thickness and ridges (EM-bird), underwater ice draft measurements from multibeam sonars mounted on submarines and AUVs. Also other field measurements of sea ice parameters have been compiled to validate ice thickness, ice types and ridges/leads that are derived from satellite radar altimeter, passive microwave data and SAR/optical data.
AWI has contributed significantly with field data from Arctic expeditions, in particular the PAMARCMIP aircraft surveys in March-April 2011 and Polarstern expedition ARKXXVI/3, in August-September 2011 (Fig. 1). Both field campaigns featured extensive measurements of sea ice thickness and freeboard. Data from many previous field experiments in sea ice areas have also been compiled and made available for the project, especially data on ice thickness, ice ridges, and other sea ice variables needed to validate the satellite retrievals. Data from several airborne EM surveys, including ice thickness and surface roughness data, as well as ground measurements are compiled and made available in the PANGAEA database (www.pangea.de) see Fig 2a.
NERSC has contributed with in situ data from a field experiments conducted in August 2011 and 2012 by KV Svalbard (Fig. 3) and NIERSC has obtained data collected from the North Pole Drifting Station in 2010-2011 and previous years. The in situ data are mainly focused on obtaining freeboard and thickness data for validation of CryoSat thickness retrievals, but the data are also used for validation of sea ice classification, ridge detection and deformation obtained from SAR data.
UCAM has provided sea ice data from two research expeditions in Fram Strait using the “Arctic Sunrise”, in September 2011 and 2012. The data collected were used to validate and calibrate satellite measurements in the marginal ice zone (MIZ) and gather high resolution data of deformed sea ice that will provide new insights into the physical processes underlying sea ice mechanics and dynamics. In order to achieve these goals we performed a variety of observations, including ice core sampling, snow depth measurements, thickness readings, aerial imagery and 3D laser scanning. UCAM has also provided submarine ULS data from 2004 and 2007 with ULS data giving ice draft measurements across the Arctic Ocean (Fig. 2b). Ice draft data is the only data type that can be used to estimate the amount of ice keels and ridges than represent the thickest ice in the Arctic Ocean. The draft data can therefore provide data on the thickness distribution, which is the most important ice thickness parameter.
WP3 Satellite data provision
Satellite data for the product development in the project have been obtained through several channels. One data source is the GMES Marine Core Services through the MyOcean project. These data are mainly large-scale modelling fields that are needed for the regional models and some satellite sea ice products provided by the Sea Ice and Wind Thematic Assembly Center in MyOcean. High-resolution SAR and optical images needed for detailed sea ice and iceberg mapping are also obtained from MyOcean and the Data Ware House (DWH). After ENVISAT stopped in April 2012, efforts have been done GMES and MyOcean to replace ENVISAT with RADARSAT2 for SAR data provision. The possibility to develop downstream services using SAR data can be affected by this change. However, the technical development of SAR-based products is not affected by changing from ENVISAT to RADARSAT2. Ocean colour data have been obtained form ENVISAT MERIS until MERIS stopped in April 2012. The work conducted on albedo and meltpond fraction is based on MERIS data from the ENVISAT period. The algorithm for retrieval of thin from L-band passive microwave data has been in progress throughout the period based on SMOS data. The radar altimeter data and derived ice thickness from CryoSat-2 has been delayed. The reason is that ESA needed to reprocess the level 1b data form CryoSat over sea ice, which implied that validation of CryoSat derived ice thickness data have not been done, but validation of the retrieval algorithms has been done. Airborne data and IceSat data have also been used for this work. Finally, ARGOS position data for tracking marine mammals in ice covered areas. Have been used. Access to test data have been obtained by agreement with users who possess data sets to be used in the project. Other satellite data needed to supplement the data mentioned above have been obtained as planned. Except for CryoSat data, the project has obtained plenty of data needed to develop and validate the satellite products.
WP4 Albedo and snow cover on sea ice
The development of a sea ice albedo and meltpond fraction algorithm based on optical satellite sensors has been the main activity in this workpackage. The algorithm uses Normalized Difference Snow Index (NDSI) to identify snow pixels. If the pixel is mapped as ‘snow’, the SGSP algorithm is used to retrieve the effective snow grain size and the soot concentration. The snow albedo is calculated with analytical formulas with given grain size and soot concentration. Pixels identifiedas as water saturated snow are discarded. If the pixel is mapped as ‘neither snow nor water saturation’, then it is considered as consisting of drained surface (white ice) and melt ponds. The melt ponds fraction is obtained using the least squares method with predefined spectral reflectance of white ice. Albedo of this pixel is calculated as a sum of the white ice albedo and melts ponds albedo with regards to angular dependences of the reflection coefficients.
In order to retrieve the albedo of sea ice, it is important to discriminate it from other surface types. Spectral resolution of the MODIS sensor makes it possible to analyze the spectral signatures of different surface types. Fig. 6a shows the difference of the bahaviour of spectral curves corresponding to snow, ice, melting ice, early and late meltponds. Vertical lines show various MODIS channels: green – band 4 (545-565nm), red – band 2 (841-876nm), brown – band 5 (1230-1250nm). It is visible that an absolute threshold on band 5 (A(5) < 0.1) is already able to discriminate between snow and ice surfaces; with the reference to bands 4 and 2 it is possible to analyze the shape of the spectral curve of a scene estimating the following ratios: (band 5/band 2) and (band 5/band 4). An example of ice albedo retrieval product is shown in Fig. 6b.
A software package has been implemented for albedo calculation and melt pond detection (MDP). This software package generates operational products in the period from spring to autumn. The software package involves processing of MERIS data from the initial data (level 1b, full orbits) till the resulting maps which contain melt pond fraction and spectral albedo for each pixel of the scene. The loss of ENVISAT in April 2012 implied that new MERIS data were not available after this date. The work has therefore been concentrated on analysis of archived MERIS data, which is most important for climate users who need longer time series. The estimation of albedo and meltpond fraction was first done for the area north of Alaska because of the availability of validation data in this area (Fig. 6). In contrast to other existing algorithms this algorithm doesn’t use a priori values of the spectral albedo of constituents of the melting ice, but is based on the original theory of the optical properties of the melting ice developed during the first year. It includes the correction of the ice characteristics and melt ponds fraction with the iterative procedure.
The original procedure for the atmospheric correction has been developed for the MPD algorithm. It is based at the suggested analytical formula that allows one to take into account the influence of the atmosphere on the satellite measured radiance coefficients. The accuracy of this formula has been investigated using the numerical modeling. A model of Arctic atmosphere to be applied to the MPD algorithm has been developed. The stability of the algorithm to the variations of the atmospheric parameters has been shown. The computer simulations have shown the required accuracy of the developed atmospheric correction procedure. MPD code including atmosphere correction procedure was delivered to Bremen University.
The available field measurements of spectral albedo (by C. Polashenski) have been processed in order to select the appropriate validation data for the retrievals developed within SIDARUS. A concept of cloud screening over snow using MODIS thermal channels has been introduced and applied to test scenes. Due to saturation problem of MODIS, it has been decided to switch the sensor and use MERIS for future work. Due to the absence of thermal channels among MERIS channels, developed cloud screening method could not be applied for MERIS sensor.
The MPD software developed by IP NANB has been incorporated into the shell-script based processing chain which utilizes MERIS L1b data available at University of Bremen. The geographical range (Beaufort sea near Point Barrow, Alaska) of MERIS scenes has been chosen to overlap with the available field data by C. Polashenski. This enabled preliminary validation and calibration of the algorithm. For the cloud clearing of MERIS data over snow, a threshold based algorithm has been suggested and implemented into the MPD software.
Analysis of the produced melt pond fraction and albedo product and comparison to the field data showed that the resulting product is in some cases contaminated by clouds which may result in overestimation of melt pond fraction and needs to be eliminated. Therefore a better method to screen out clouds has been suggested. It utilizes the oxygen-A absorption band of MERIS, which is sensitive to the cloud height. Unfortunately, the shift of MERIS pixels along the wavelength – the “smile” effect – is amplified within the oxygen-A band, so that the existing “smile” correction is not sufficient. Radiative transfer modeling was used to model the transmittance of each pixel and to calculate the correction factor, which enables better smile correction and accurate cloud clearing. The advantage of this method is its sensitivity to high clouds, which are a challenging case for the threshold-based algorithm.
The original procedure for the atmospheric correction has been developed for the MPD algorithm. It is based at the suggested analytical formula that allows one to take into account the influence of the atmosphere on the satellite measured radiance coefficients. The accuracy of this formula has been investigated using the numerical modeling. A model of Arctic atmosphere to be applied to the MPD algorithm has been developed. The stability of the algorithm to the variations of the atmospheric parameters has been shown. The computer simulations have shown the required accuracy of the developed atmospheric correction procedure. MPD code including atmosphere correction procedure was delivered to Bremen University.
The available field measurements of spectral albedo (by C. Polashenski) have been processed in order to select the appropriate validation data for the retrievals developed within SIDARUS. A concept of cloud screening over snow using MODIS thermal channels has been introduced and applied to test scenes. Due to saturation problem of MODIS, it has been decided to switch the sensor and use MERIS for future work. Due to the absence of thermal channels among MERIS channels, developed cloud screening method could not be applied for MERIS sensor.
The MPD software developed by IP NANB has been incorporated into the shell-script based processing chain which utilizes MERIS L1b data available at University of Bremen. The geographical range (Beaufort sea near Point Barrow, Alaska) of MERIS scenes has been chosen to overlap with the available field data by C. Polashenski. This enabled preliminary validation and calibration of the algorithm. For the cloud clearing of MERIS data over snow, a threshold based algorithm has been suggested and implemented into the MPD software.
Analysis of the produced melt pond fraction and albedo product and comparison to the field data showed that the resulting product is in some cases contaminated by clouds which may result in overestimation of melt pond fraction and needs to be eliminated. Therefore a better method to screen out clouds has been suggested. It utilizes the oxygen-A absorption band of MERIS, which is sensitive to the cloud height. Unfortunately, the shift of MERIS pixels along the wavelength – the “smile” effect – is amplified within the oxygen-A band, so that the existing “smile” correction is not sufficient. Radiative transfer modeling was used to model the transmittance of each pixel and to calculate the correction factor, which enables better smile correction and accurate cloud clearing. The advantage of this method is its sensitivity to high clouds, which are a challenging case for the threshold-based algorithm.
Conclusions of the work are the following:
- Global Arctic product of sea ice albedo and melt pond fraction are produced for the summer season using the NSIDC grid 12.5km.
- MERIS data have been processed for the whole summer of 2009 and for June 2002-2011 (time span of entire MERIS dataset).
- Weekly averaged products remove most of the cloud coveage and show pond evolution during the summer season
- Trends of MPF show melt onset starting somewhat earlier on FYI and also on MYI
- Variety of factors determining surface albedo cannot be reduced to air temperature at the surface
- Comparison of meltpond product by study by Roesel et al. shows a discrepancy which need further investigation
WP5 SAR analysis for sea ice and icebergs
The development of SAR-based sea ice and iceberg products were focussed on (1) sea ice classification, (2) iceberg detection and tracking, and (3) sea ice motion and deformation analyses. The responsible researchers for those tasks were from NIERSC and NERSC (classification), CLS (icebergs), and AWI (drift).
In the first phase, activities in this work package were focused on a review of SAR algorithms available for the three tasks listed above, and on developing a concept and a prototype for a “toolbox”. The latter is a collection of special algorithms for classification, target detection, and target tracking that is optimized for handling. The review of different SAR algorithms was published as Deliverable 5.1 (D5.1) (Review of SAR algorithm and toolbox). The concept for a toolbox is summarized in Deliverable 5.2 (D5.2).
For SAR sea ice classification using single polarization ENVISAT data, two approaches were regarded as optimal, namely the Bayesian classification (BC), and the Neural Network (NN) approach. The former is based on calculating the probability that a certain ice type is present at the location of interest, and on minimizing the error probability. In the NN approach, the input information (in this case radar intensity and image texture parameters) is fed through a network of decision rules that fix how to separate different ice types. The network has to be “trained” by an interactive procedure in which the output (the final classification) is provided manually by an experienced ice analyst, using suitable SAR images or parts of single images. Examples were provided for sea ice classification in the central Arctic. Here, both approaches recognized the major ice types. The products from the pixel-based BC approach were more noisy but provided a better detection of narrow leads. The NN-classification was less noisy. Another example focused on the application of the NN-approach to the marginal ice zone, using a combination of SAR images and passive microwave data. The results of the NN-based classification were published in an article in IEEE Trans. Of Geoscience and Remote Sensing (Zakhvatkina et al., 2012).
After ENVISAT stopped in April 2012, the work was shifted to use of dual polarization Radarsat-2 data. These data enabled further improvement of SAR ice classification, where both polarizations are used in the algorithm. For Radarsat-2 data the Neural Network (NN) method was replaced by the Support Vector Machine (SVM) method. The other steps of the processing steps are unchanged (SAR pre-processing, selection of ice types and training areas, running of the classification, and validation of the results (Fig. 10 and 11.). The polarizations are HH- and HV. The latter reveals noisy patterns that are removed before classification. The sea ice classification approach is based on the Support Vector Machines (SVM) approach, which is a supervised learning method. Besides radar intensities, eight textural parameters were selected as most informative parameters using statistical analysis. The texture characteristics considered are correlation, inertia, cluster prominence, energy, homogeneity, and entropy, as well as 3rd and 4th central statistical moments of image brightness. To include the effect of wind in the classification procedure, it is distinguished between calm and rough water surfaces.
Results showed that the separation of water and ice is easier when both polarizations are available. A clear advantage of the dual-polarization mode for discriminating different ice types is not found for the ice conditions at the selected test site (Fram Strait). On the basis of single-polarization Envisat ASAR images acquired during winter in the central Arctic, the performance of Bayesian and Neural Network classification was assessed in comparison to ice charts and interpretations of sea-ice mapping specialists. It was found that in many cases multi-year, rough first-year, smooth first-year, and new ice could be distinguished with varying accuracies. The achievable correspondence between the results of the algorithms and the complementary data sets was from 68 to 91% for multi-year ice and from 83 to 99% for first-year ice (including smooth and rough types) in the case of the Neural Network approach, and between 40 and 90% for multi-year ice and from 88 to 99% for first-year ice in case of the Bayesian algorithm.
For sea ice drift, cross-correlation schemes, feature matching, and the optical flow method are introduced. Different pre-processing steps such as geocoding and image enhancement are described that are necessary for drift retrieval. For the SIDARUS-project, the use of a cascaded multi-resolution approach for drift calculation is recommended. For sea ice drift and deformation monitoring, an algorithm was implemented and tested using Envisat single-polarization images from a polynia region in the Weddell Sea (Antarctic) and dual-polarization ScanSAR images from Radarsat-2, acquired over Fram Strait (Fig. 12).
The algorithm requires stable ice structures that can be identified in two consecutive SAR images. This condition is usually fulfilled in the pack ice with its prominent floes and deformation zones. A procedure was developed to assess the reliability of the retrieved ice drift patterns if no additional buoy data are available. It is found that the drift fields derived from HH- and HV-polarized images differ slightly since the sensitivity to certain ice structures depends on polarization. This item needs to be further addressed.
Validate of the quality of the motion tracking result in the absence of reference data is to do a a simple consistency check called backmatching [e.g. Schreer, 2005]. Here, the displacement field is calculated twice, with an interchanged image sequence. In regions where the algorithm makes “bad guesses” due to insufficient texture, the results from both runs are inconsistent. This method is able to identify those highly textured but very dynamic regions the CFA cannot discern (Fig. 13). Examples for such regions are found, for instance, in polynias and the ice margin zone.
Fig. 13 shows displacement fields calculated from a Radarsat-2 image pair recorded on the 16.09.2012 with ≈10h time difference between image acquisitions. The map on the left shows the displacement field estimated from HH polarization, and the image on the right shows the result from the HV-polarized data. The red arrows are visually determined reference data. From a comparison between backmatching and texture analysis, we can discern the following regions:
❶ In this region, displacement vectors are marked as reliable by both approaches.
❷ This region is marked as unreliable by both the texture-based approach and backmatching. In the example above, this is mostly open water, but also homogeneous-looking areas
-Here, the texture-based approach considers an area to be reliable which is found to be inconsistent by backmatching. This happens, for instance, at the ice margin zone with its fast-changing ice conditions.
❹ Regions that fall into this class are considered reliable by backmatching, but fail the texture-based quality criterion.
It can be seen in Fig. 13 that there are slightly different regions marked as reliable, depending on polarization. The displacement field calculated from HH-polarized SAR images shows fewer gaps than the result obtained from cross-polarized data. It has been indicated that both channels are sensitive to different structures in the ice and hence contain complementary information [Komarov, 2012], and our own analysis supports this finding.
The discussion on the implementation of the toolbox (D5.2) revealed that the optimal solution was to develop three different toolboxes: one for classification, one for ice drift detection, and one for ice drift and deformation retrieval. Reasons were (a) that each group already had software available, which included modules that could not freely be shared among partners, (b) groups used different programming environments, (c) a separate implementation increased the flexibility in development by the different partners, and also the flexibility in application and testing by different end-users.
Drift and deformation: The retrieval of sea ice drift and deformation from SAR images is based on a cascaded multi-scale technique comprising phase correlation and normalized cross correlation. The algorithm for retrieving the displacement vector between the different positions of an ice surface structure that can be recognized in two images uses the input images processed at different spatial resolutions (“resolution pyramid”), gradually refining the resulting drift field (“cascade”). A major challenge is to provide information on the reliability of the drift field. A robust possibility to achieve this is to carry out the calculations two times, the second time using image #2 as starting point (backmatching). This approach, however, doubles the computation time. Therefore, another measure is introduced that combines the analysis of textural and correlation parameters in a confidence factor (CFA). The CFA takes into account six parameters. Three of them are from the field of image texture analysis, namely the mean intensity gradient (MIG), the mean gradient slope (MGS), and the variance-to-squared-mean ratio (VMR). The fourth is simply an intensity threshold that reduces the effect of mirror reflections from the ice. These four parameters characterize the properties of each individual image but do not reflect any links between a pair of images. Therefore correlation coefficient and its confidence interval are included in the reliability assessment. Sea ice deformation parameters are calculated using the drift field as input. As output, the total deformation rate, convergence/divergence, and vorticity are obtained.
Deliverable 5.4 (D5.4) was the completed toolbox. However, because of the conclusions resulting from the work on D5.2 three separate toolbox modules were implemented. It was decided to supplement the software by a description of the structure and handling of each module, which is provided as additional report. D5.4 was due after two and a half year of project runtime. However, work on the optimization of the software codes was continued.
The final status and results of WP 5 are presented in Deliverable 5.5 (D5.5). The developed software modules were applied to different data. The background of the selected algorithms is provided as well as examples of applications. In the following, the final status of the three tasks is provided:
ICEBERG DETECTION
Concerning iceberg detection in SAR images, different approaches have been investigated, such as Constant False Alarm Rate (CFAR) methods, detection of iceberg shadows, edge detection and segmentation, or Optimization Polarimetric Contrast Enhancement (OPCE). Iceberg detection using SAR and altimeter is established in the Antarctica where large icebergs are relatively easy to detect (Fig. 14). In the Arctic, icebergs are much smaller and more difficult to detect with existing satellite sensors, especially in the Barents and Kara Seas. The report D5.1 provides pros and cons of the different methods and suggests possible improvements., e. g. by also considering information on the iceberg drift.
The detection of smaller icebergs that drift in the open ocean or are trapped in sea ice has been investigated. For this purpose, Radarsat-2 ScanSAR-Wide (spatial resolution 50 m) and ScanSAR-Narrow (25m) images were analysed. Results presented in D5.3 were that detections from HV-polarized images are less reliable and that a spatial resolution of 25m increases the rate of reliable detections considerably. To improve the identification of icebergs in sea ice, initial tests with different methods for image segmentation were carried out. The MDL (minimum description length) algorithm is found to be most promising for sea ice segmentation, but a number of problems have still to be solved in practical applications.
WP6 Sea ice thickness
Thin ice from SMOS data
Two retrieval algorithms for thickness of thin sea ice from passive microwave data from SMOS data have been tested, one based on intensity and incidence angles below 40°, and one based on simultaneously using both intensity and polarization difference at incidence angles above 40°. The algorithms have been implemented as a prototype product, which will be important for mapping of thin ice areas in both Arctic and Antarctic (Fig. 16). Studies are ongoing where time series of thin ice are analyzed and compared with other data and modeling results. A main activity in the future will be to carry out validation activities, because there is very little in situ data on thin ice. Data on thin ice will be very important because more of the Arctic sea ice is firstyear ice, while in the Antarctic the ice cover is dominated by is firstyear ice. The ice thickness data from SMOS can be used to quantify the areas freezing from summer to winter.
There is presently very little data on thin ice that can be used to validate sea ice models. Compilation of in situ ice thickness data has continued, which is required for validation of CryoSat ice thickness retrievals.. It is primarily in situ data from multiyear ice that is required because CryoSat will obtain freeboard data for ice thicker than 1 m. It is also important to obtain data on ridge distribution, which has been done by UCAM in recent field experiments. The ice thickness from SMOS has been included into the operational ice service at met.no.
Ice thickness retrieval algorithm for CryoSat data
The aim of this study is to develop, validate and select algorithm for SIT retrieval from CryoSat2. The new developed A(FD2) algorithm for CryoSat2 is validated with collocated SID from ULS and SIT from laser altimeter (LA) on board Operational Ice Bridge (OIB) by comparison of SIT and SID derived from ULS and LA with collocated SID data. The CryoSat2 A(FD2) algorithm with minimum bias is selected. The accuracy of the FD algorithm is confirmed by comparison of SID and SIT derived from collocated moored and on Submarine ULS, LA and RA. ESA/CryoSat2, NSIDC, climate change, cryosphere and numerical prediction models will benefit the results of this study.
Assuming hydrostatic equilibrium, the SIT, hi, can be retrieved from the freeboard, hfi, measured from CryoSat2 by:
hi=(hsrs +hfirw)/(rw - ri), (1)
where the snow depth (hs) and density (rs) from Warren climatology [Warren, 1999] (WC) as a function of latitude, longitude and month of the year in the Arctic have been used until now [Laxon et al, 2012].
Assuming that the radar returns are from the snow–ice interface, which is valid for low temperature and dry snow, the SID, retrieved from RA is calculated as:
dra=hi-hfi = (hsrs +hfiri)/( rw - ri), (2)
where hi is the SIT, calculated by Equation 1, and hfi is the retrieved freeboard from radar altimeter. Water density, rw, and ice density, ri, depend on temperature, salinity and ice type, but to simplify the algorithm, constant values have been used from different authors, leading to incompatible results and errors in estimated SIT from RA [Connor et al, 2009, Laxon et al, 2012]. Seven algorithms for freeboard to SIT conversion have been compared, validated and the impact of sea ice density, snow depth density and water density has been examined [Djepa and Wadhams, 2013]. The validation (with ULS and OIB/laser altimeter) and sensitivity analyses demonstrated that the assumption of the half snow depth over first year ice (FYI) and fixed ice densities over FYI and MYI will lead to underestimation of the SIT.
The SIT, retrieved from the airborne laser altimeter (LA/ATM) on board Operational Ice Bridge (OIB), has been used for algorithm selection and validation. The SIT, retrieved from LA [Kurtz et al 2012] is snow depth, density, ice density and freeboard dependent as the SIT retrieved from RA and is calculated from the freeboard retrieved from the airborne laser scanner (hf) at the air-snow interface by:
hi= rwhf/(rw-ri)-( r w-rs)hs/(rw-ri) (3)
where (hs), is the snow depth, ri , rs , rw are the ice, snow and water densities. The laser altimeter measures the freeboard on air snow interface, hf, and the radar altimeter measure the freeboard on ice snow interface in the presence of dry snow and cold conditions (Figure 17).
The CryoSat retrieval algorithm has been investigated and improved after comparison with a number of other data obtained from submarine profiles, Upward-Looking Sonar (ULS) moorings, AUVs, airborne data and other. A sensitivity analysis was carried out to quantify how changes in ice density and snow depth has impact on sea ice thickness (SIT) retrievals from freeboard measurements, using the hydrostatic equilibrium equation. The results of this sensitivity analysis is shown in Fig. 18.
Validation of ice thickness retrievals from radar altimeter data from ERS and ENVISAT
Since CryoSat data has been delayed, very limited studies have been possible to carry out before the end of the project. Therefore, more work has been focussed on studying thickness retrievals from the radar altimeter data from ERS and ENVISAT. A number of examples of comparison of between thickness retrieval from radar altimeter and other thickness data are presented in the following figures
Preliminary studies of CryoSat data
A case study of CryoSat data has been conducted in the Fram Strait. Met.no has extracted relevant data from CryoSat-2 to determine how to improve its capability to detect sea ice thickness by comparing it with available in situ measurements, Synthetic Aperture Radar (SAR) data, and sea ice charts as dependent variables for ice thickness proxies. The following describes the background and application by Met.no for each data source.
Average waveforms from the level 1 data were used to determine if a criteria can be established in which radar altimetry measurements can detect sea ice thickness variations based on measurements from the waveform amplitudes. Cryosat-2 SAR mode level 1b data waveforms were converted to power in Watts with knowledge of the scale factor and power. Characteristic waveforms over sea ice show rougher signatures due to the irregularity of surface features. However, indicative patterns that include new thin ice or leads should display these features as having the highest amplitudes, whereas smaller waveforms represent surface roughness. Depending on where these open water or thin ice areas occur within the waveform, this information can theoretically be used to infer sea ice thickness. Waveforms are clearly defined to determine whether the surface is ocean or sea ice (Figure 24). Indicative features show a small peak prior to a dramatic larger peak due to noise from the reflection of surrounding elevated features. Several passes were combined and overlain on to the Radarsat-2 SAR data to illustrate how well it can detect areas of open water (Figure 25). (http://www.altimetry.info/html/use_cases/data_use_case_cryosat_2-2_en.html).
WP7 Forecasting of sea ice and icebergs
The sea ice forecasting system for the Barents and Kara Seas has been developed to produce forecasts of ice and ocean variables every week for up to 7 days. The system is based on the TOPAZ model and data assimilation system developed at NERSC (TP4), which is the main forecasting system for the Arctic region within the MyOcean project. The TOPAZ ice-ocean data and model system is the combination of the Hybrid Coordinate Ocean Model (HYCOM, Bleck 2002, and the Ensemble Kalman Filter (EnKF, Evensen, 2009) and covers the Nordic and Arctic Seas at a horizontal resolution of about 12-16 km (Bertino and Lisæter, 2008, Sakov et al. 2012, see http://topaz.nersc.no). The models use 28 hybrid z-isopycnal layers in the vertical. The TOPAZ system is today operational at the Norwegian Meteorological Institute (met.no) and daily forecast and download of data are available within the MYOCEAN platform (myocean.met.no/ARC-MFC).
The regional model over the Barents and Kara Seas (BS1) covers the area of the Barents Sea and the Kara Sea as well as a some parts of the areas in the Fram Strait, Nordic Seas, and Arctic Ocean, as shown in Fig. 27.
A high-resolution 510x450 grid is applied, that gives approximately a 5 km horizontal resolution. Nesting conditions, taken from the assimilated TOPAZ system, and tidal forcing are applied at the open boundaries. River (TRIP) and atmospheric forcing (ECMWF) are applied.
The applied sea ice model in the NERSC HYCOM model system distinguishes between the sea ice rheology in the consolidated ice pack and in the marginal ice zone (MIZ). The rheology used today in the consolidated ice pack is an elastic-viscous-plastic (EVP) model based on Hunke and Dukowicz 1997. A new MIZ rheology has been implemented into the NERSC HYCOM model system (Shen et al, 1987), where internal stresses are based on statistics of chaotic collisions between circular floes of a given size. The rheology has been further developed to include the mechanical breaking of ice floes using the mechanical parameters of sea ice (Dumont et al, 2011).
A wave-in-ice module (WIM) has been implemented into the system, where surface waves are propagated into ice-covered areas and break up the ice into smaller flows (Williams et al, 2012a, 2013a,b). The WIM model propagates wave characteristics, taken from and external wave forecast model, into the ice and keeps wave characteristics in the model memory. Surface waves impose strain in the ice and may break up the ice into smaller floes. The waves dissipate energy travelling into the ice, due to internal ice resistance and wave reflection, and will at some position be too weak to break up the ice. With the WIM model we are able to define a new criteria for the MIZ based on the floe size, here set to 200 m. In the present setup wave data is taken from the 10 km WAM North Sea forecast model (WAMNSEA10km) operated by the Meteorology Institute in Norway. A nearest point solution is applied to convert the wave characteristics onto the model grid.
The impact of the WIM module are clearly seen on the 26th of Mars 2013, when waves with large amplitude reach the ice edge, see Figure 28 below. The waves taken from an external model reach the ice edge (subplot c) and break up the ice into smaller floes (subplot d) south of Svalbard. The criterion for the MIZ rheology is set to dfloe<200m, so a new large area where maximum floe size is less than 200m are define as MIZ area, and will then apply the MIZ type of sea ice rheology.
The BS1 model runs on a daily basis producing 3 days forecast of sea ice and ocean conditions. The nesting cycle and the forecast procedure are described in Figure 29. The figure shows the weekly forecast cycle starting at a Tuesday, here described as Day0. After one week on the next Tuesday, here Day+7, the cycle repeats itself. The forecast system runs, and is dependent on forecast products on several computer servers, AT NERSC. The main server in the forecast system is HEXAGON, the supercomputer system at the University of Bergen, from where the forecast system is initiated by running shell scripts using job scheduler Crontab functions. The scripts starts to download forecast products; atmospheric forecast ECMWFR from VILJE (ntnu.no) and wave forecast WAMNNSEA10km, TP4 restart files as initial conditions, and OSI-SAF sea ice concentration from the ftp server at myocean.met.no. The outer model, TP4, is running ones a week, giving nesting condition to the inner model, BS1. The inner model runs on a daily basis using daily updated forecast products. When the BS1 forecast model is ready, forecast and validation figures are produced and send over to the NANSEN server at nersc.no. The figures are sorted and a script is initiated that update the web page at the NANSEN server. The system is dependent on four different servers and four different forecast products, see Figure 29.
A webpage is established and maintained where the daily sea ice forecast, for the next three proceeding days, see EVP+MIZ below. An automatic system is setup that downloads forcing fields, validation data, run the TP4 and the BS1 model, produce figures, and update the webpage. The webpage also include a Validation section where earlier forecasts are compared to OSI-SAF sea ice concentration fields. A parallel forecast system and webpage are set up for comparison; presenting results using the newly developed wave-in-ice module, see EVP+MIZ+WIM below. The WIM forecast is only for the next two proceeding days, due to the limitation of the + 60h surface wave forecast.
• EVP+MIZ: http://topaz.nersc.no/Knut/IceForecast/Barents
• EVP+MIZ+WIM: http://topaz.nersc.no/Knut/IceForecast/Barents2
WP8 Data integration and validation
This section describes how various SIDARUS products have been validated and used together with other data from the project.
Validation of ice-water discrimination from SAR
Sea ice classification results calculated using SVM technique have been compared with Met.no (Norwegian Meteorological Institute) ice charts from the operational ice charting service. Validation of Arctic ice products is always a challenging task due to lack of ground truth data. As a substitute, our product has been intercompared with a manual sea ice product produced by Met.no.
Met.no ice charts are produced by ice analysts at the Norwegian Sea Ice Service using the following data sources: high resolution microwave Synthetic Aperture Radar data (Radarsat), low resolution microwave SSM/I and SSMIS data (DMSP), MODIS data (Terra and Aqua) and AVHRR data from NOAA.
The procedure of Met.no and SVM classification comparison is illustrated by Fig. 30. Met.no sea ice type data has been reclassified into charts with only three classes: open water (OW), where sea ice concentration values on the original ice map were in the range from 0 to 15% , sea ice, where sea ice concentration values were in the range from 15% to 100% and land. Our results of SVM classification have been reclassified likewise - into three classes - OW, ice and land. In the comparison Met.no ice charts have been assumed to represent correct classification and we calculated the confusion matrix (error matrix) keeping in mind that assumption.
The overall accuracies of SVM classification were estimated for about 776 Radarsat-2 SAR images received in the period from March – November 2013. The average values of total accuracies for each month are presented in the table below. The accuracy is the ratio of the number of similarly classified pixels on both (Met.no and SVM) ice maps to the whole number of ice and water pixels in the image. The error is the ratio of the number of incorrectly classified pixels of each class to the total number of pixels of that class.
The average accuracy of our sea ice classification for the period March-November 2013 is 0.89 being higher in winter (0.92) and lower in summer (0.86). The accuracy is lower in summer months because the SVM classification algorithm was tuned using winter data.
Below an example of comparison of automated SAR classification and Met.no ice charts are presented. The images are shown in groups. Every group consists of two satellite images and two ice charts: Radarsat-2 HH image and Terra Modis image represent the ice conditions in the scene from the microwave and visual point of view, and two ice charts - the result of our SVM classification and the Met.no sea ice chart - demonstrate how ice conditions are described by automated classification and sea ice analysts. Additionally the confusion matrix is given for every particular pair of ice charts (SVM's result vs. Met.no). In the matrix the correctness of classification for different classes based on pixel-by-pixel differences is shown. The structure of the accuracy/error matrix is the following table:
Validation of ice drift and deformation from SAR
The validation of ice drift fields retrieved from satellite imagery can be carried out using drifting buoys. However, the number of buoys deployed in Arctic or Antarctic waters is small, and in a lot of scenes, a buoy may not be present at all. But even if this is the case, only a very limited area in the satellite-derived drift map is covered by the buoy track. Another well-suited possibility is the manual generation of a reference drift field by an experienced operator, which is then used for comparison with the automatically retrieved result. This procedure can only be carried out for a few examples but not for every image pair used in operational mapping. In the framework of the SIDARUS project, a proxy for the accuracy was developed that provides information whether the retrieved drift vectors are reliable. A robust method is to carry out the drift calculations two times, the second time using image 2 as starting point (back-matching). This approach, however, doubles the computation time. Therefore, another measure is introduced in addition that combines the analysis of textural and correlation parameters in a confidence factor (CFA). The CFA takes into account six parameters. Three of them are from the field of image texture analysis, namely the mean intensity gradient (MIG), the mean gradient slope (MGS), and the variance-to-squared-mean ratio (VMR). The fourth is simply an intensity threshold that reduces the effect of mirror reflections from the ice. These four parameters characterize the properties of each individual image but do not reflect any links between a pair of images. Therefore the correlation coefficient and its confidence interval are included in the reliability assessment. A result of such a reliability assessment is shown in Fig. 32. Sea ice deformation parameters are calculated using the drift field as input, hence their accuracy depends on the accuracy of the drift vectors. Also the ratio between spatial extension of the deformation structure and the grid used to calculate the drift, as well as the spatial resolution of the SAR images, have an influence on the uncertainties in the deformation parameters. Quantitative studies of the accuracy of different deformation parameters were not planned for SIDARUS but are recently carried out as a follow-on investigation.
Validation of the Barents – Kara Sea model
For every day the model is run, a validation is performed where the first day forecast four days earlier of sea ice concentration is compared to OSI-SAF sea ice concentrations. The 15% OSI-SAF concentration line is plotted on top of the model sea ice concentration and on top of other forecast products.
Further validation has been done off-line. In general does the forecast model follow the seasonal variability and respond correct to atmospheric forcing, though without direct assimilation within the regional model, some major deviations from measurements are seen. I) The transition from no ice to high sea ice concentration is to sharp in the model compared to OSI-SAF data, see Figure 33 a) and b). II) To much sea ice is seen north of Svalbard and west of Novaya Zemlya during most of the winter season. III) Even in sea ice thickness does the model have a too sharp transition from thin towards thicker sea ice compared to SMOS sea ice thickness, see Figure 33 c) and d).
Validation of thin ice from SMOS
Validation of thin ice is an important but complicated issue, since it is too thin to stand or walk on it. Therefore the validation of a thin ice thickness retrieval has to rely on other remote sensing data. One source of thin ice data was a night time thermal imagery sea ice thickness retrieval from the Moderate-resolution Imaging Spectroradiometer (MODIS). For Winter 2010/11 71 MODIS scenes are analyzed and compared with the SMOS retrieval showing good agreement especially in lower ice thicknesses as can be seen from Fig. 34.
Another source of validation Data was the AWI EM-bird instrument which can be carried by a helicopter or airplane which measures every few meters the ice thickness of a footprint of about 50m diameter. Fig. 35 shows one EM bird flight track and the corresponding SMOS sea ice thickness retrieval. The logarithmic histograms on the left show the distribution of EM bird measured thicknesses with their mean, median, and the SMOS retrieved values at that location number from the right-hand side. Here it is visible that the SMOS retrieval is in nearly all cases exactly at the peak occurrence of EM-bird measured sea ice thickness.
The validation shows that the sea ice thickness retrieval from SMOS works within the indicated limits.
Validation of CryoSat ice thickness
For validation of the CryoSat ice thickness data we have used a set of Radarsat-2 data in Fine Quad-Pole SLC and SGF beam mode. The satellite data have been acquired to coincide with in-situ measurements from two cruises in the Arctic: KV Svalbard (6-13 April, 2011) and Arctic Sunrise (7-12 July, 2012). Fine quad-pol Radarsat-2 SAR data was used as a ground-truth proxy for sea ice type when compared with CryoSat 2 level 1 waveforms. The CryoSat 2 SIRAL instrument is programmed to operate in SAR mode over sea ice areas, as its measurement footprint has a reduced surface area that allows the detection of smaller sea ice features. Though the use of level 2 CryoSat-2 data was preferred for this comparison due to the inclusion of multiple parameters (i.e. retracker, sea surface height, freeboard, elevation..etc.),investigation of the available products found that these fields had not been calculated. Therefore, we extracted the average waveforms from the level 1 data and converted them to power in Watts with knowledge of the scale factor and power.
Characteristic waveforms over sea ice can theoretically be used to infer sea ice thickness depending on where open water or thin ice areas occur within the waveform. However, in order for these waveforms to accurately depict the surface roughness it is necessary to implement the appropriate retracking algorithm. The following retracking algorithms are currently available:
• UCL: ESA retracker
• AWI: Threshold-Spline-Retracker Algorithm
• NOAA: Ocean height based on Maximum Likelihood Estimator
• FMI: Open water and new ice threshold retracker with a Gaussian and Gaussian + exponential fit
• Traditional OCOG retracker
• Primary peak OCOG retracker
Though several retrackers have been used with previous corrections, specific conditions require different methods of fitting the tracking point on the leading edge and the algorithms vary with each mode. The level 2 data implemented an Offset Centre-of-Gravity (OCOG) and an OCOG threshold retracker but requires further evaluation to resolve errors in the return.
However, in order for these waveforms to accurately depict the surface roughness it is necessary to implement the appropriate retracking algorithm to determine at which point the waveform is actually measuring the surface from a nadir view rather than showing effects of related to noise from how the signal varies in the range direction. Though several retrackers have been used with previous corrections, specific conditions require different methods of fitting the tracking point on the leading edge and the algorithms vary with each mode. The level 2 data implemented an OCOG and an OCOG threshold retracker but requires further evaluation to resolve errors in the return. Therefore, it will be necessary for Met.no to customize thresholds and parameters to the level 1b data to fit the needs of doing a robust comparison with SAR and in situ data.
Validation of ice albedo and meltponds
The Melt Pond Detection (MPD) algorithm has been validated using a dataset of validation data shown in Table 8.1. The data can be divided into two categories: in situ data (Fig. 36) and airborne data (Fig. 36). For each category of validation data, a collocation and comparison of retrieved and field value has been performed. For the in situ data, the correlation coefficient is 0.526. The observed scatter can be connected to low quality of some validation data (visual estimation of melt pond fraction) and different spatial resolution of the retrieval and the validation data. Here only one example of airborne validation is shown (Fig. 36), whereas the whole dataset comprises 10 comparison cases. The correspondence of the MPD retrieved values to the airborne data is reasonably good (40% field value against 50% satellite) with several exceptions which can be explained due to inability of MERIS to resolve the difference between melt pond and so called “blue ice” (sea ice without top scattering layer).
The validation of the MPD retrieval against field data is a challenging task due to lack of validation data which can be collocated with the MERIS data spatially and temporally without being sorted out due to cloud coverage. Another possibility to validate the MPD retrieval is comparison to another remote sensing product. Such comparison attempt has been performed (not shown here) and showed some discrepancies between the MPD retrieval and algorithm by Roesel et al., 2011, from MODIS data. Due to different overflight times of the two sensors (MODIS and MERIS), the cloud coverage and surface coverage might have been slightly different. On the scale of weeks (weekly averages have been compared), this could have caused some discrepancy. In addition, the difference in accounting for surface reflectance within both retrievals could have been the reason for the disagreement. Further comparison is needed to clarify the issue.
Potential Impact:
The results of SIDARUS are expected to have impact on climate research as well as marine operations and environmental protection in polar regions. The Arctic and Antarctic is dominated by ice-covered oceans and coasts. The regions are exposed to climate change with significant impact on the cryosphere and the environment that is affected by the presence of ice. In the Arctic the global warming is at roughly twice the global average rate, with a dramatic reduction in summer sea ice extent as one of the clearest indicators of this trend. Physical and biological processes are being transformed across the entire regions while climate feedback mechanisms in the Arctic’s changing atmospheric and oceanic dynamics impact at global scales. The Arctic regions offer vast areas of hydrocarbon resources that have just started to be exploited. The Arctic Ocean is surrounded by continental shelves, where in particular the huge Siberian shelf covering the eastern hemisphere, extending from the Barents Sea to the Chukchi Sea.
The ongoing changes in Arctic climate with increasing temperatures and decreasing sea ice cover have stimulated the interest for oil and gas exploration in several Arctic areas. A reduction of the sea ice area opens up the possibility to access new areas of the Arctic Ocean where hydrocarbon resources can be exploited and transported to the markets. The main Arctic areas where large-scale offshore exploration have started are: Sakhalin in Sea of Okhotsk, North Slope of Alaska, Cook Inlet, Grand Banks of Newfoundland, Barents Sea (Snøhvit field and the upcoming Shtokman field) and the Pechora Sea. A map of regions in the Arctic with potential oil and gas fields is shown in Fig. 1.
All these areas have seasonal sea ice cover and some have icebergs that put severe constraints on design and operation of installations and on transport solutions. Even if the sea ice cover is decreasing and is expected to diminish further in the coming decades, the sea ice will still remain a dominant factor in most of the exploration areas in the winter season. In the summer months, however, less sea ice will provide access to offshore areas in Canada, Greenland and on the eastern Siberian shelf that were previously inaccessible due to sea ice. Improved sea ice end iceberg monitoring and forecasting in the polar regions will have large socio-economic impact due to climate change combined with increased human activities. The global demand for energy has stimulted the political interest for the Arctic Ocean and several countries have started investigations of the continental shelves. Under the UN Convention on the Law of the Sea, a country can claim exclusive economic rights within 200 miles (Fig.2) If a country can prove that its continental shelf extends beyond the 200-mile economic zone, it can claim similar rights over a larger area. All States involved in the Arctic Ocean continental shelf have ratified the Convention except the USA.
Ice information products, such as charts and forecasts, are provided by the national meteorological or oceanographic services of countries with activities in ice-affected waters. Currently, shipping is the primary user of these products (Fig. 3). They are created by combining data from satellites, in situ sensors, and aerial and shipboard observations. Each source has strengths and weaknesses. In situ sensors, aerial surveys and ships provide specific but sparse information, and aerial surveys are expensive. Satellite data are not as detailed, but they are systematic, cost-effective and cover wide areas. A variety of satellite sensors provide data at varying resolutions, spatial scales and costs. Because ice can be highly dynamic, ice products must be synthesised quickly (in 1-6 hours) and regularly (every 6-24 hours, every day). Synthesis of data archives allows for statistical analysis and prediction.
The results of SIDARUS will improve sea ice information and be useful for offshore industry and shipping polar areas, scientists working in climate and environmental research and organisations working with protection of sea ice habitats. The existing GMES provide services large-scale sea ice concentration and ice drift maps from satellite data and ice-ocean forecasting for the whole Arctic with about 20 km grid cells. The forecasting products are provided by the TOPAZ system at NERSC (http://topaz.nersc.no). Large scale sea ice charts are provided by University of Bremen (http://iup.physik.uni-bremen.de:8084/amsr/amsre.html) the ice drift maps are produced by Ifremer (http://cersat.ifremer.fr/data/discovery) and regional high-resolution ice charts for the Svalbard area by Met Norway. Other products cover Greenland waters, Canadian areas and Antarctica.
SIDARUS will provide local ice information for areas around planned drilling areas and along sailing routes, using high resolution SAR and optical satellite images, with pixel size ≈ 10 m or smaller. The products from the satellite images will include maps of ice concentration, ice drift and ice type classification, the latter including a separating of smooth level and deformed ice (ridges, shear zones), open water, leads, and areas of thin ice. Satellite data alone cannot provide all required information about sea ice and icebergs. Therefore, SIDARUS will also include additional data, in particular ice thickness using airborne surveys with the HEM system and in situ observations with the EM31 instrument and with multibeam sonar under the ice. SIDARUS will also provide sea ice and iceberg forecasting for the Barents Sea, using high-resolution models nested to the TOPAZ system, and a web-based ice information system where new as well as archived data and model simulations are made available. This will enable users to obtain both statistical information based on previous observations and model simulations, new observations based on multisensory approach, and short-term (up to 10 days) forecasting using NCEP forcing fields. The services developed in SIDARUS will first be implemented for the Barents Sea region, with focus on the Shtokman area that will start to be developed in 2012. The services will later be implemented in other areas where offshore industry plan to start operations. In the Antarctic, iceberg monitoring and forecasting will be implemented. The services will complement and extend the present GMES Marine Core services.
There is a growing interest for combined information on sea ice mapping and animal tracking provided, for example for polar bears, as expressed by Dr Andrew Rocher, Professor at University of Alberta, and Chair of the IUCN/SSC Polar Bear Specialist Group: “The remoteness of polar bear habitat, the low density of bears, and the long periods of winter darkness preclude most field research methods. Satellite telemetry has thus played a major role in developing insights into their ecology and aiding their management, dating back to the 1980s”. High resolution products provided by SIDARUS would also help the wildlife user group: Earth Observation data at medium resolution (12.5 km) such as AMSR-E passive microwave instrument to know the actual sea ice situation. For example, the Alaska Science Centre of the US Geological Survey tracked walruses in Summer 2007 and found them in open water in end of July. They presumed that there were remnant ice floes that the walruses were using as haulout platforms. However this high resolution information was not available from the medium resolution products. Examples of user organisations working with protection of sea ice habitats are: (1) The Institute of Evolution and Ecology Problems (Russian Academy of Sciences). They are already using satellite data (Argos) for wildlife surveillance and they are keen on this new concept involving ice data. They are very interested in ice density and the breathing facility of mammals under the ice. This would be a totally new way of understanding wildlife behavior. (2) The Alberta University, Canada. The Department of Biological Sciences has completed long term studies on Ecology, conservation, and management of large Arctic mammals focusing on polar bears. Interests centre on limiting and regulating factors of polar bear populations including habitat use, harvest effects, and predator-prey relationships. Current research includes assessment of the effects of climate change on polar bears. That University has already worked on joint projects with the Wegner Institute in Germany. (3) The Greenland Institute of Natural Resources, Denmark. The Greenland Institute is ready to share its humpty whales tracking data in Iceland. The Institute plans to study narwhales In East Greenland during the summer 2010. These mammals will presumably move around the sea ice in Greenland and their movements will be correlated with ice information from SIDARUS. Also, the Institute is tagging polar bears with Argos collars between Canada and Greenland.
The Intergovernmental Panel on Climate Change (IPCC) has provided several scenarios for future climate change in different parts of the world [5]. For the Arctic, the observed trend of reduced ice extent will continue and the summer ice extent may disappear towards the end of the century. The record low ice extent in September 2007 and 2012 suggested that the summer ice may disappear much sooner (Fig.4). In the early 1980s the summer ice area was about 6.5 mill km2, and it diminished by about 8 % per decade until 2006. In 2007 it the area shrunk to less than 4.5 mill km2. The reduction of ice area of more than 2 mill km2 corresponds to the size of Greenland. When the summer ice disappears, the thick multiyear ice will be absent, leaving the Arctic Ocean with a thinner ice cover in the winter season. The disappearance of multiyear ice combined with longer melt season will have important operational implications, leading to greater access and longer navigation season for shipping around the Arctic basin.
The IPCC scenarios suggest that there will be a substantial warming in Arctic and sub-Arctic areas compared to the present situation, which is closely linked to the reduction of the sea ice. The surface air temperature in many climate model projections shows a 6-8°C warming over the ocean during winter, with a less dramatic change in terrestrial regions. With higher temperature and reduction of the ice cover, the marginal ice zone will move poleward, leaving the coastal and shelf areas ice-free in the summer.
With a smaller area covered by sea ice, more heat from solar radiation will be absorbed in the ocean, leading to increased ocean temperature. A warmer ocean will in turn reduce the amount of sea ice formed in the following winter. This is a so-called positive feedback mechanism, leading to enhanced warming in the Arctic. Another effect of a warmer Arctic is more clouds and precipitation. Increased fog will result sin in poorer surface visibility, which is an obstacle for many operations. More frequent and stronger storms can also be expected in the sub-Arctic areas. Vessel icing could also increase in these areas, especially during outbreak of cold Arctic continental air masses.
How can we assess the climate change impact in the various sub-Arctic regions where offshore operations are foreseen ? During winter, the central Arctic and all peripheral seas including the Greenland Sea, Bering Sea, and Gulf of St. Lawrence will continue to have significant ice cover. Ice extent and thickness will generally be reduced. The Sea of Okhotsk and Sea of Japan will be ice-free for the entire year. In late summer, the entire Russian coast will be ice free, allowing navigation through the Barents, Kara, Laptev and East Siberian Seas along the entire Northern Sea Route [6]. This situation has already been observed in the last couple of summers. The Northwest Passage through the Canadian Archipelago and along the coast of Alaska will in general be ice free and navigable in summer by non-icebreaking ships. Ice will be present all year along the eastern and northern coasts of Greenland. Ice will also remain throughout the summer within and adjacent to the northern Canadian Archipelago. However, severe winters with more ice than average may also be expected due to the natural variability of the climate system. The effect of more wind and waves in ice-covered areas will be increased ridging and stamukhas in near coastal regions. The iceberg situation in different parts of the Arctic is difficult to assess, but it is likely that more icebergs can occur in some years as a consequence of diminishing Arctic glaciers. Arctic shipping is expected to increase as a consequence of less sea ice and more offshore exploration.
The possible consequences of increased oil and gas exploration in the vulnerable Arctic environment is a controversial issue. The Arctic ecosystems are already today exposed to severe treats due to the effects of a warmer climate. The climate effect comes in addition to the latent risk of radioactive contamination due to extensive nuclear bomb testing in the Russian Arctic in the previous decades. The storage of nuclear waste from scrapped reactors is also a severe risk factor, because it is not clear how safe this storage will be in the future. A growing oil and gas industry operation on land as well as at sea will increase the pressure on the environment with increased risk of accidents that can have severe and long-lasting negative effects on ecosystems. A worst-case scenario is an Exxon-Valdez type of accident that occurred in Alaska in 1989. The ecosystems in the area affected by this accident are still marked by this oil pollution disaster, almost 20 years after it happened [8]. The environmental impact of oil and gas exploration will be higher in the Arctic compared to other areas in the world. This calls for new technologies to ensure safe operations as well as legislative norms that regulate the activities. These factors are not in place yet and need to be developed.
In conclusion, offshore operations in the Arctic will be more feasible as a consequence of the climate change, leading to less sea ice and warmer temperatures, The costs of operations, however, will be high due to extreme ice and weather conditions and requirements to operate with minimum risks to harm the vulnerable Arctic environment. This requires that adequate ice monitoring and forecasting systems are developed, validated and implemented. SIDARUS results can make a significant contribution to this goal.
List of Websites:
http://sidarus.nersc.no
Prof. Stein Sandven
Director
Nansen Environmental and Remote Sensing Center (NERSC)
Tel: +47 55 20 58 00
Fax: +47 55 20 58 01
E-mail: stein.sandven@nersc.no
The objective of SIDARUS has been to develop new sea ice data products from satellite, airborne and underwater platforms as well as ice forecasting from models in order to improve sea ice monitoring services. In addition to analysis of satellite and airborne data, the project has analyzed in situ, airborne and under-ice data from previous and new field campaigns. These are essential data for validation of satellite retrievals of sea ice parameters before the satellite products can be implemented as services for users. The results of SIDARUS are obtained in the following six categories: (1) SAR based products on regional and local scale, where new algorithms for ice classification, ice edge detection, ice drift, ice deformation and iceberg detection have been developed. The SAR data have been obtained from ENVISAT, Radarsat-2 and TerraSAR-X under agreement with GMES. The main application of these products is to improve the operational ice monitoring services; (2) A sea ice albedo retrieval algorithm based on optical satellite data, mainly MERIS data form ENVISAT, has been developed. The algorithm has been implemented to generate maps of sea ice albedo and meltpond fraction in the period from spring to autumn. The main application of these products will be in atmosphere-ice-ocean modelling. The main users group is therefore sea ice and climate modellers who need improved parametrization of albedo and meltpond fraction in their models; (3) Sea ice thickness analysis has included a retrieval algorithm for thin ice (< 0.5 m) based on 1.4 GHz passive microwave data from the recently launched ESA SMOS mission. Measurement of thin ice is a very innovative method, which will be supplementary to the thickness retrievals from altimeter data from IceSat and CryoSat. Altimeter measurement of sea ice freeboard can be used to quantify the thicker part of the ice cover, typically multiyear ice and firstyear ice thicker than about 1 m. In order to validate the ice thickness retrievals, several non-satellite methods have been used. These include profiles of submarine, AUV and mooring-based upward-looking sonar data, electromagnetic induction measurements from helicopters, and in situ data from field experiments. Studies have been conducted to validate thickness retrievals from altimeters, but there is need for considerable more validation work. Ice thickness data are needed by many users, ranging from sea ice and climate modelers to operational ice services, ice navigators, polar shipping and offshore industry working in the Arctic. (4) An ice-ocean forecasting model on regional scale has implemented in the Barents and Kara Sea with improved representation of sea ice dynamics in the marginal ice zone. This model is will support offshore operators and ships that need better ice forecasting in this region. SIDARUS has also supported development of a commercial iceberg service for ships in the Antarctica. The service combines observed icebergs from satellite data with a regional Antarctic circumpolar current model in order to provide forecasts of icebergs. The main customer for this service so far is the Vendee Global race, but other ships operating in Antarctica can also benefit form the service. (5) SIDARUS has demonstrated how ARGOS position data from tagged animals can be combined with regional sea ice maps, where SAR data have been used. These products are important for studying the behaviour of marine mammals in the polar regions. The main user group are environmental monitoring and wildlife protection institutions. (6) Several field campaigns in the Arctic have been supporter by SIDARUS in order to improve collection of validation data for the satellite retrievals. The campaigns have provided data from in situ, underwater and airborne platforms, in particular for ice thickness estimation.
The data sets and products have been demonstrated to users working in climate research, marine safety and environmental monitoring. The results of SIDARUS will be used to extend the present GMES marine services with new satellite-derived sea ice products, ice forecasting from regional models and validation of sea ice products using non-satellite data. The demand for improved sea ice information in the Arctic and Antarctic by many user groups is growing as a result of climate change and its impact on environment and human activities. The presently observed reduction of the Arctic sea ice extent, in particular during the summer months and an increasing demand for natural resources are key mechanisms driving human activities in these areas.
Project Context and Objectives:
Context of SIDARUS
The demand for sea ice information in Arctic and Antarctic seas by many user groups is growing as a result of climate change and its impact on environment and human activities. The presently observed reduction of the Arctic sea ice extent, in particular during the summer months and an increasing demand for natural resources are key mechanisms driving human activities in these areas. The EU has started to develop a policy for the Arctic where protection of the environment and sustainable use of resources are issues of high priority. The expected growth in ship traffic, oil and gas exploration, fisheries and tourism in the coming years will increase the risk for accidents affecting environment, health, safety and economy in this unique and vulnerable region.
Sea ice is a crucial component of the climate system, influenced by conditions in both the atmosphere and ocean. Variations in sea ice may in turn modulate climate by altering the surface albedo; the exchange of heat, moisture, and momentum between the atmosphere and ocean; and the upper ocean stratification in areas of deep-water formation. Recent reductions and changes in the Arctic sea ice cover have already shown severe impact on the environment and climate in the Arctic, with predictions that the changes will accelerate in the next decades. The climate change in the Arctic, and its impact on human activities and environment, has generated increasing demand for sea ice information on past, present and future scenarios. Sea ice data is required by several user groups: marine transportation operators and regulators, meteorological organizations, Search and Rescue organizations, defence and security agencies, resource developers (oil, gas, timber, minerals, fish), policy-makers, marine engineers, populations living and working in ice-affected regions, and environmental institutions working with sea ice habitat protection. Last but not least, climate modellers need significantly improved sea ice representation in the climate models.
Sea ice observation from satellites has been done for more than three decades. Several sensors and retrieval methods have been developed and successfully utilized to measure sea ice area, concentration and drift (e.g. Breivik et al., 2009). Today operational sea ice monitoring and analysis is fully dependent on use of satellite data. However, new and improved satellite systems require further development of sea ice remote sensing. Monitoring of sea ice concentration, extent, and drift using satellite data is provided by the Marine Core services of MyOcean and other services. Use of SAR data for operational sea ice monitoring has been introduced in MyOcean to produce regional ice charts and ice drift maps (http://www.myocean.eu.org). In the GSE project PolarView, several sea ice services are provided, as described at http://www.polarview.org/services/. SIDARUS will provide new sea ice products for climate, marine safety and protection of sea ice habitats as well as in situ data for validation of satellite-derived sea ice products.
Sea ice monitoring and other metocean services (meteorology, oceanography and sea ice) are presently providing large- and regional-scale ice charts and forecasting for the Arctic and Antarctic seas. GMES ice services deliver satellite- and model-based products that aim at general applications for a wide range of users. However, high latitude users such as offshore industry and Arctic-Antarctic shipping require more specific services in addition to the present GMES services. For climate research and prediction, there is a set of essential climate variables (ECVs) where satellites can provide important data. In addition to ice area and ice drift, which are already monitored by existing GMES services, data on sea ice thickness, leads/polynyas, surface albedo, and snow cover are needed, but these variables are not included in the services delivered by MyOcean or other GMES projects. SIDARUS has de eloped new sea ice service products including high-resolution sea ice data to support operators on regional and local scale as well as large scale information for climate monitoring. SIDARUS has developed SAR-based products as preparation for operational use of Sentinel-1 data for sea ice and iceberg monitoring. Furthermore, SIDARUS has provided validation data for satellite-retrieved sea ice products, and implemented regional ice forecasting using high-resolution ice –ocean models in order to meet the requirements from offshore operators and shipping activities in different regions of the Arctic.
Objectives of SIDARUS
The following specific objectives of SIDARUS have been to
(1) Develop sea ice classification and iceberg detection using new high-resolution SAR images with different frequency and polarization, and implement a monitoring service based on SAR data from Sentinel-1
(2) Provide sea ice thickness data for thin ice (<≈ 0.5 m) using the new 1.4 GHz passive microwave data from SMOS, as a complement to the ice thickness data from CryoSat
(3) Collect and analyze data on sea ice thickness and other ice parameters data from airborne, in situ and underwater platform experiments in order to validate satellite retrievals and fill gaps in sea ice observations that satellite data cannot provide
(4) Develop and validate sea ice albedo and meltpond retrieval from multi-spectral optical images, e.g MODIS on the EOS platforms and MERIS and AATSR on ENVISAT
(5) Provide integrated maps of marine mammal tracks from ARGOSS data and sea ice maps from satellite data
(6) Implement a high-resolution operational ice-ocean model in order to provide sea ice and iceberg forecasts on regional and local scale
(7) Demonstrate sea ice monitoring and forecasting services to user groups by integration of observational products from several platforms and simulation/forecasting products from global and regional models using GIS and web technology
(8) Plan sustainable sea ice downstream services for GMES, consisting of both public domain and commercial products
Project Results:
WP1 User interaction
The consortium has been in contact with users of the sea ice products and services throughout the project period. The users include representatives for each of the five application areas where products are being developed: (1) High-resolution sea ice and iceberg mapping by SAR to support marine transportation, offshore operations, and ice-weather forecasting services in polar regions. (2) Sea ice albedo from optical sensors to support sea ice modelling and climate research, (3) Sea ice thickness from satellite data, airborne surveys and other observing platforms providing data for climate research and operational users. (4) Ice forecasting based on numerical models and satellite data for marine safety. (5) ARGOS tracking of marine mammals combined with sea ice maps to support environmental management, wildlife protection. Products in all these five categories have been developed and disseminated to users through workshops and conferences. In addition the consortium members have had meetings with identified users throughout the project period. In some cases users have directly supported the downstream service development through addition projects, supplementing the work in SIDARUS.
WP2 Field campaigns with data collection
Sea ice data from several field campaigns have been obtained in the period. In particular in situ ice thickness data have been collected because these data are important for validation of the satellite retrievals from satellite altimeter data. The ice thickness data used in the project includes airborne electromagnetic measurements of ice thickness and ridges (EM-bird), underwater ice draft measurements from multibeam sonars mounted on submarines and AUVs. Also other field measurements of sea ice parameters have been compiled to validate ice thickness, ice types and ridges/leads that are derived from satellite radar altimeter, passive microwave data and SAR/optical data.
AWI has contributed significantly with field data from Arctic expeditions, in particular the PAMARCMIP aircraft surveys in March-April 2011 and Polarstern expedition ARKXXVI/3, in August-September 2011 (Fig. 1). Both field campaigns featured extensive measurements of sea ice thickness and freeboard. Data from many previous field experiments in sea ice areas have also been compiled and made available for the project, especially data on ice thickness, ice ridges, and other sea ice variables needed to validate the satellite retrievals. Data from several airborne EM surveys, including ice thickness and surface roughness data, as well as ground measurements are compiled and made available in the PANGAEA database (www.pangea.de) see Fig 2a.
NERSC has contributed with in situ data from a field experiments conducted in August 2011 and 2012 by KV Svalbard (Fig. 3) and NIERSC has obtained data collected from the North Pole Drifting Station in 2010-2011 and previous years. The in situ data are mainly focused on obtaining freeboard and thickness data for validation of CryoSat thickness retrievals, but the data are also used for validation of sea ice classification, ridge detection and deformation obtained from SAR data.
UCAM has provided sea ice data from two research expeditions in Fram Strait using the “Arctic Sunrise”, in September 2011 and 2012. The data collected were used to validate and calibrate satellite measurements in the marginal ice zone (MIZ) and gather high resolution data of deformed sea ice that will provide new insights into the physical processes underlying sea ice mechanics and dynamics. In order to achieve these goals we performed a variety of observations, including ice core sampling, snow depth measurements, thickness readings, aerial imagery and 3D laser scanning. UCAM has also provided submarine ULS data from 2004 and 2007 with ULS data giving ice draft measurements across the Arctic Ocean (Fig. 2b). Ice draft data is the only data type that can be used to estimate the amount of ice keels and ridges than represent the thickest ice in the Arctic Ocean. The draft data can therefore provide data on the thickness distribution, which is the most important ice thickness parameter.
WP3 Satellite data provision
Satellite data for the product development in the project have been obtained through several channels. One data source is the GMES Marine Core Services through the MyOcean project. These data are mainly large-scale modelling fields that are needed for the regional models and some satellite sea ice products provided by the Sea Ice and Wind Thematic Assembly Center in MyOcean. High-resolution SAR and optical images needed for detailed sea ice and iceberg mapping are also obtained from MyOcean and the Data Ware House (DWH). After ENVISAT stopped in April 2012, efforts have been done GMES and MyOcean to replace ENVISAT with RADARSAT2 for SAR data provision. The possibility to develop downstream services using SAR data can be affected by this change. However, the technical development of SAR-based products is not affected by changing from ENVISAT to RADARSAT2. Ocean colour data have been obtained form ENVISAT MERIS until MERIS stopped in April 2012. The work conducted on albedo and meltpond fraction is based on MERIS data from the ENVISAT period. The algorithm for retrieval of thin from L-band passive microwave data has been in progress throughout the period based on SMOS data. The radar altimeter data and derived ice thickness from CryoSat-2 has been delayed. The reason is that ESA needed to reprocess the level 1b data form CryoSat over sea ice, which implied that validation of CryoSat derived ice thickness data have not been done, but validation of the retrieval algorithms has been done. Airborne data and IceSat data have also been used for this work. Finally, ARGOS position data for tracking marine mammals in ice covered areas. Have been used. Access to test data have been obtained by agreement with users who possess data sets to be used in the project. Other satellite data needed to supplement the data mentioned above have been obtained as planned. Except for CryoSat data, the project has obtained plenty of data needed to develop and validate the satellite products.
WP4 Albedo and snow cover on sea ice
The development of a sea ice albedo and meltpond fraction algorithm based on optical satellite sensors has been the main activity in this workpackage. The algorithm uses Normalized Difference Snow Index (NDSI) to identify snow pixels. If the pixel is mapped as ‘snow’, the SGSP algorithm is used to retrieve the effective snow grain size and the soot concentration. The snow albedo is calculated with analytical formulas with given grain size and soot concentration. Pixels identifiedas as water saturated snow are discarded. If the pixel is mapped as ‘neither snow nor water saturation’, then it is considered as consisting of drained surface (white ice) and melt ponds. The melt ponds fraction is obtained using the least squares method with predefined spectral reflectance of white ice. Albedo of this pixel is calculated as a sum of the white ice albedo and melts ponds albedo with regards to angular dependences of the reflection coefficients.
In order to retrieve the albedo of sea ice, it is important to discriminate it from other surface types. Spectral resolution of the MODIS sensor makes it possible to analyze the spectral signatures of different surface types. Fig. 6a shows the difference of the bahaviour of spectral curves corresponding to snow, ice, melting ice, early and late meltponds. Vertical lines show various MODIS channels: green – band 4 (545-565nm), red – band 2 (841-876nm), brown – band 5 (1230-1250nm). It is visible that an absolute threshold on band 5 (A(5) < 0.1) is already able to discriminate between snow and ice surfaces; with the reference to bands 4 and 2 it is possible to analyze the shape of the spectral curve of a scene estimating the following ratios: (band 5/band 2) and (band 5/band 4). An example of ice albedo retrieval product is shown in Fig. 6b.
A software package has been implemented for albedo calculation and melt pond detection (MDP). This software package generates operational products in the period from spring to autumn. The software package involves processing of MERIS data from the initial data (level 1b, full orbits) till the resulting maps which contain melt pond fraction and spectral albedo for each pixel of the scene. The loss of ENVISAT in April 2012 implied that new MERIS data were not available after this date. The work has therefore been concentrated on analysis of archived MERIS data, which is most important for climate users who need longer time series. The estimation of albedo and meltpond fraction was first done for the area north of Alaska because of the availability of validation data in this area (Fig. 6). In contrast to other existing algorithms this algorithm doesn’t use a priori values of the spectral albedo of constituents of the melting ice, but is based on the original theory of the optical properties of the melting ice developed during the first year. It includes the correction of the ice characteristics and melt ponds fraction with the iterative procedure.
The original procedure for the atmospheric correction has been developed for the MPD algorithm. It is based at the suggested analytical formula that allows one to take into account the influence of the atmosphere on the satellite measured radiance coefficients. The accuracy of this formula has been investigated using the numerical modeling. A model of Arctic atmosphere to be applied to the MPD algorithm has been developed. The stability of the algorithm to the variations of the atmospheric parameters has been shown. The computer simulations have shown the required accuracy of the developed atmospheric correction procedure. MPD code including atmosphere correction procedure was delivered to Bremen University.
The available field measurements of spectral albedo (by C. Polashenski) have been processed in order to select the appropriate validation data for the retrievals developed within SIDARUS. A concept of cloud screening over snow using MODIS thermal channels has been introduced and applied to test scenes. Due to saturation problem of MODIS, it has been decided to switch the sensor and use MERIS for future work. Due to the absence of thermal channels among MERIS channels, developed cloud screening method could not be applied for MERIS sensor.
The MPD software developed by IP NANB has been incorporated into the shell-script based processing chain which utilizes MERIS L1b data available at University of Bremen. The geographical range (Beaufort sea near Point Barrow, Alaska) of MERIS scenes has been chosen to overlap with the available field data by C. Polashenski. This enabled preliminary validation and calibration of the algorithm. For the cloud clearing of MERIS data over snow, a threshold based algorithm has been suggested and implemented into the MPD software.
Analysis of the produced melt pond fraction and albedo product and comparison to the field data showed that the resulting product is in some cases contaminated by clouds which may result in overestimation of melt pond fraction and needs to be eliminated. Therefore a better method to screen out clouds has been suggested. It utilizes the oxygen-A absorption band of MERIS, which is sensitive to the cloud height. Unfortunately, the shift of MERIS pixels along the wavelength – the “smile” effect – is amplified within the oxygen-A band, so that the existing “smile” correction is not sufficient. Radiative transfer modeling was used to model the transmittance of each pixel and to calculate the correction factor, which enables better smile correction and accurate cloud clearing. The advantage of this method is its sensitivity to high clouds, which are a challenging case for the threshold-based algorithm.
The original procedure for the atmospheric correction has been developed for the MPD algorithm. It is based at the suggested analytical formula that allows one to take into account the influence of the atmosphere on the satellite measured radiance coefficients. The accuracy of this formula has been investigated using the numerical modeling. A model of Arctic atmosphere to be applied to the MPD algorithm has been developed. The stability of the algorithm to the variations of the atmospheric parameters has been shown. The computer simulations have shown the required accuracy of the developed atmospheric correction procedure. MPD code including atmosphere correction procedure was delivered to Bremen University.
The available field measurements of spectral albedo (by C. Polashenski) have been processed in order to select the appropriate validation data for the retrievals developed within SIDARUS. A concept of cloud screening over snow using MODIS thermal channels has been introduced and applied to test scenes. Due to saturation problem of MODIS, it has been decided to switch the sensor and use MERIS for future work. Due to the absence of thermal channels among MERIS channels, developed cloud screening method could not be applied for MERIS sensor.
The MPD software developed by IP NANB has been incorporated into the shell-script based processing chain which utilizes MERIS L1b data available at University of Bremen. The geographical range (Beaufort sea near Point Barrow, Alaska) of MERIS scenes has been chosen to overlap with the available field data by C. Polashenski. This enabled preliminary validation and calibration of the algorithm. For the cloud clearing of MERIS data over snow, a threshold based algorithm has been suggested and implemented into the MPD software.
Analysis of the produced melt pond fraction and albedo product and comparison to the field data showed that the resulting product is in some cases contaminated by clouds which may result in overestimation of melt pond fraction and needs to be eliminated. Therefore a better method to screen out clouds has been suggested. It utilizes the oxygen-A absorption band of MERIS, which is sensitive to the cloud height. Unfortunately, the shift of MERIS pixels along the wavelength – the “smile” effect – is amplified within the oxygen-A band, so that the existing “smile” correction is not sufficient. Radiative transfer modeling was used to model the transmittance of each pixel and to calculate the correction factor, which enables better smile correction and accurate cloud clearing. The advantage of this method is its sensitivity to high clouds, which are a challenging case for the threshold-based algorithm.
Conclusions of the work are the following:
- Global Arctic product of sea ice albedo and melt pond fraction are produced for the summer season using the NSIDC grid 12.5km.
- MERIS data have been processed for the whole summer of 2009 and for June 2002-2011 (time span of entire MERIS dataset).
- Weekly averaged products remove most of the cloud coveage and show pond evolution during the summer season
- Trends of MPF show melt onset starting somewhat earlier on FYI and also on MYI
- Variety of factors determining surface albedo cannot be reduced to air temperature at the surface
- Comparison of meltpond product by study by Roesel et al. shows a discrepancy which need further investigation
WP5 SAR analysis for sea ice and icebergs
The development of SAR-based sea ice and iceberg products were focussed on (1) sea ice classification, (2) iceberg detection and tracking, and (3) sea ice motion and deformation analyses. The responsible researchers for those tasks were from NIERSC and NERSC (classification), CLS (icebergs), and AWI (drift).
In the first phase, activities in this work package were focused on a review of SAR algorithms available for the three tasks listed above, and on developing a concept and a prototype for a “toolbox”. The latter is a collection of special algorithms for classification, target detection, and target tracking that is optimized for handling. The review of different SAR algorithms was published as Deliverable 5.1 (D5.1) (Review of SAR algorithm and toolbox). The concept for a toolbox is summarized in Deliverable 5.2 (D5.2).
For SAR sea ice classification using single polarization ENVISAT data, two approaches were regarded as optimal, namely the Bayesian classification (BC), and the Neural Network (NN) approach. The former is based on calculating the probability that a certain ice type is present at the location of interest, and on minimizing the error probability. In the NN approach, the input information (in this case radar intensity and image texture parameters) is fed through a network of decision rules that fix how to separate different ice types. The network has to be “trained” by an interactive procedure in which the output (the final classification) is provided manually by an experienced ice analyst, using suitable SAR images or parts of single images. Examples were provided for sea ice classification in the central Arctic. Here, both approaches recognized the major ice types. The products from the pixel-based BC approach were more noisy but provided a better detection of narrow leads. The NN-classification was less noisy. Another example focused on the application of the NN-approach to the marginal ice zone, using a combination of SAR images and passive microwave data. The results of the NN-based classification were published in an article in IEEE Trans. Of Geoscience and Remote Sensing (Zakhvatkina et al., 2012).
After ENVISAT stopped in April 2012, the work was shifted to use of dual polarization Radarsat-2 data. These data enabled further improvement of SAR ice classification, where both polarizations are used in the algorithm. For Radarsat-2 data the Neural Network (NN) method was replaced by the Support Vector Machine (SVM) method. The other steps of the processing steps are unchanged (SAR pre-processing, selection of ice types and training areas, running of the classification, and validation of the results (Fig. 10 and 11.). The polarizations are HH- and HV. The latter reveals noisy patterns that are removed before classification. The sea ice classification approach is based on the Support Vector Machines (SVM) approach, which is a supervised learning method. Besides radar intensities, eight textural parameters were selected as most informative parameters using statistical analysis. The texture characteristics considered are correlation, inertia, cluster prominence, energy, homogeneity, and entropy, as well as 3rd and 4th central statistical moments of image brightness. To include the effect of wind in the classification procedure, it is distinguished between calm and rough water surfaces.
Results showed that the separation of water and ice is easier when both polarizations are available. A clear advantage of the dual-polarization mode for discriminating different ice types is not found for the ice conditions at the selected test site (Fram Strait). On the basis of single-polarization Envisat ASAR images acquired during winter in the central Arctic, the performance of Bayesian and Neural Network classification was assessed in comparison to ice charts and interpretations of sea-ice mapping specialists. It was found that in many cases multi-year, rough first-year, smooth first-year, and new ice could be distinguished with varying accuracies. The achievable correspondence between the results of the algorithms and the complementary data sets was from 68 to 91% for multi-year ice and from 83 to 99% for first-year ice (including smooth and rough types) in the case of the Neural Network approach, and between 40 and 90% for multi-year ice and from 88 to 99% for first-year ice in case of the Bayesian algorithm.
For sea ice drift, cross-correlation schemes, feature matching, and the optical flow method are introduced. Different pre-processing steps such as geocoding and image enhancement are described that are necessary for drift retrieval. For the SIDARUS-project, the use of a cascaded multi-resolution approach for drift calculation is recommended. For sea ice drift and deformation monitoring, an algorithm was implemented and tested using Envisat single-polarization images from a polynia region in the Weddell Sea (Antarctic) and dual-polarization ScanSAR images from Radarsat-2, acquired over Fram Strait (Fig. 12).
The algorithm requires stable ice structures that can be identified in two consecutive SAR images. This condition is usually fulfilled in the pack ice with its prominent floes and deformation zones. A procedure was developed to assess the reliability of the retrieved ice drift patterns if no additional buoy data are available. It is found that the drift fields derived from HH- and HV-polarized images differ slightly since the sensitivity to certain ice structures depends on polarization. This item needs to be further addressed.
Validate of the quality of the motion tracking result in the absence of reference data is to do a a simple consistency check called backmatching [e.g. Schreer, 2005]. Here, the displacement field is calculated twice, with an interchanged image sequence. In regions where the algorithm makes “bad guesses” due to insufficient texture, the results from both runs are inconsistent. This method is able to identify those highly textured but very dynamic regions the CFA cannot discern (Fig. 13). Examples for such regions are found, for instance, in polynias and the ice margin zone.
Fig. 13 shows displacement fields calculated from a Radarsat-2 image pair recorded on the 16.09.2012 with ≈10h time difference between image acquisitions. The map on the left shows the displacement field estimated from HH polarization, and the image on the right shows the result from the HV-polarized data. The red arrows are visually determined reference data. From a comparison between backmatching and texture analysis, we can discern the following regions:
❶ In this region, displacement vectors are marked as reliable by both approaches.
❷ This region is marked as unreliable by both the texture-based approach and backmatching. In the example above, this is mostly open water, but also homogeneous-looking areas
-Here, the texture-based approach considers an area to be reliable which is found to be inconsistent by backmatching. This happens, for instance, at the ice margin zone with its fast-changing ice conditions.
❹ Regions that fall into this class are considered reliable by backmatching, but fail the texture-based quality criterion.
It can be seen in Fig. 13 that there are slightly different regions marked as reliable, depending on polarization. The displacement field calculated from HH-polarized SAR images shows fewer gaps than the result obtained from cross-polarized data. It has been indicated that both channels are sensitive to different structures in the ice and hence contain complementary information [Komarov, 2012], and our own analysis supports this finding.
The discussion on the implementation of the toolbox (D5.2) revealed that the optimal solution was to develop three different toolboxes: one for classification, one for ice drift detection, and one for ice drift and deformation retrieval. Reasons were (a) that each group already had software available, which included modules that could not freely be shared among partners, (b) groups used different programming environments, (c) a separate implementation increased the flexibility in development by the different partners, and also the flexibility in application and testing by different end-users.
Drift and deformation: The retrieval of sea ice drift and deformation from SAR images is based on a cascaded multi-scale technique comprising phase correlation and normalized cross correlation. The algorithm for retrieving the displacement vector between the different positions of an ice surface structure that can be recognized in two images uses the input images processed at different spatial resolutions (“resolution pyramid”), gradually refining the resulting drift field (“cascade”). A major challenge is to provide information on the reliability of the drift field. A robust possibility to achieve this is to carry out the calculations two times, the second time using image #2 as starting point (backmatching). This approach, however, doubles the computation time. Therefore, another measure is introduced that combines the analysis of textural and correlation parameters in a confidence factor (CFA). The CFA takes into account six parameters. Three of them are from the field of image texture analysis, namely the mean intensity gradient (MIG), the mean gradient slope (MGS), and the variance-to-squared-mean ratio (VMR). The fourth is simply an intensity threshold that reduces the effect of mirror reflections from the ice. These four parameters characterize the properties of each individual image but do not reflect any links between a pair of images. Therefore correlation coefficient and its confidence interval are included in the reliability assessment. Sea ice deformation parameters are calculated using the drift field as input. As output, the total deformation rate, convergence/divergence, and vorticity are obtained.
Deliverable 5.4 (D5.4) was the completed toolbox. However, because of the conclusions resulting from the work on D5.2 three separate toolbox modules were implemented. It was decided to supplement the software by a description of the structure and handling of each module, which is provided as additional report. D5.4 was due after two and a half year of project runtime. However, work on the optimization of the software codes was continued.
The final status and results of WP 5 are presented in Deliverable 5.5 (D5.5). The developed software modules were applied to different data. The background of the selected algorithms is provided as well as examples of applications. In the following, the final status of the three tasks is provided:
ICEBERG DETECTION
Concerning iceberg detection in SAR images, different approaches have been investigated, such as Constant False Alarm Rate (CFAR) methods, detection of iceberg shadows, edge detection and segmentation, or Optimization Polarimetric Contrast Enhancement (OPCE). Iceberg detection using SAR and altimeter is established in the Antarctica where large icebergs are relatively easy to detect (Fig. 14). In the Arctic, icebergs are much smaller and more difficult to detect with existing satellite sensors, especially in the Barents and Kara Seas. The report D5.1 provides pros and cons of the different methods and suggests possible improvements., e. g. by also considering information on the iceberg drift.
The detection of smaller icebergs that drift in the open ocean or are trapped in sea ice has been investigated. For this purpose, Radarsat-2 ScanSAR-Wide (spatial resolution 50 m) and ScanSAR-Narrow (25m) images were analysed. Results presented in D5.3 were that detections from HV-polarized images are less reliable and that a spatial resolution of 25m increases the rate of reliable detections considerably. To improve the identification of icebergs in sea ice, initial tests with different methods for image segmentation were carried out. The MDL (minimum description length) algorithm is found to be most promising for sea ice segmentation, but a number of problems have still to be solved in practical applications.
WP6 Sea ice thickness
Thin ice from SMOS data
Two retrieval algorithms for thickness of thin sea ice from passive microwave data from SMOS data have been tested, one based on intensity and incidence angles below 40°, and one based on simultaneously using both intensity and polarization difference at incidence angles above 40°. The algorithms have been implemented as a prototype product, which will be important for mapping of thin ice areas in both Arctic and Antarctic (Fig. 16). Studies are ongoing where time series of thin ice are analyzed and compared with other data and modeling results. A main activity in the future will be to carry out validation activities, because there is very little in situ data on thin ice. Data on thin ice will be very important because more of the Arctic sea ice is firstyear ice, while in the Antarctic the ice cover is dominated by is firstyear ice. The ice thickness data from SMOS can be used to quantify the areas freezing from summer to winter.
There is presently very little data on thin ice that can be used to validate sea ice models. Compilation of in situ ice thickness data has continued, which is required for validation of CryoSat ice thickness retrievals.. It is primarily in situ data from multiyear ice that is required because CryoSat will obtain freeboard data for ice thicker than 1 m. It is also important to obtain data on ridge distribution, which has been done by UCAM in recent field experiments. The ice thickness from SMOS has been included into the operational ice service at met.no.
Ice thickness retrieval algorithm for CryoSat data
The aim of this study is to develop, validate and select algorithm for SIT retrieval from CryoSat2. The new developed A(FD2) algorithm for CryoSat2 is validated with collocated SID from ULS and SIT from laser altimeter (LA) on board Operational Ice Bridge (OIB) by comparison of SIT and SID derived from ULS and LA with collocated SID data. The CryoSat2 A(FD2) algorithm with minimum bias is selected. The accuracy of the FD algorithm is confirmed by comparison of SID and SIT derived from collocated moored and on Submarine ULS, LA and RA. ESA/CryoSat2, NSIDC, climate change, cryosphere and numerical prediction models will benefit the results of this study.
Assuming hydrostatic equilibrium, the SIT, hi, can be retrieved from the freeboard, hfi, measured from CryoSat2 by:
hi=(hsrs +hfirw)/(rw - ri), (1)
where the snow depth (hs) and density (rs) from Warren climatology [Warren, 1999] (WC) as a function of latitude, longitude and month of the year in the Arctic have been used until now [Laxon et al, 2012].
Assuming that the radar returns are from the snow–ice interface, which is valid for low temperature and dry snow, the SID, retrieved from RA is calculated as:
dra=hi-hfi = (hsrs +hfiri)/( rw - ri), (2)
where hi is the SIT, calculated by Equation 1, and hfi is the retrieved freeboard from radar altimeter. Water density, rw, and ice density, ri, depend on temperature, salinity and ice type, but to simplify the algorithm, constant values have been used from different authors, leading to incompatible results and errors in estimated SIT from RA [Connor et al, 2009, Laxon et al, 2012]. Seven algorithms for freeboard to SIT conversion have been compared, validated and the impact of sea ice density, snow depth density and water density has been examined [Djepa and Wadhams, 2013]. The validation (with ULS and OIB/laser altimeter) and sensitivity analyses demonstrated that the assumption of the half snow depth over first year ice (FYI) and fixed ice densities over FYI and MYI will lead to underestimation of the SIT.
The SIT, retrieved from the airborne laser altimeter (LA/ATM) on board Operational Ice Bridge (OIB), has been used for algorithm selection and validation. The SIT, retrieved from LA [Kurtz et al 2012] is snow depth, density, ice density and freeboard dependent as the SIT retrieved from RA and is calculated from the freeboard retrieved from the airborne laser scanner (hf) at the air-snow interface by:
hi= rwhf/(rw-ri)-( r w-rs)hs/(rw-ri) (3)
where (hs), is the snow depth, ri , rs , rw are the ice, snow and water densities. The laser altimeter measures the freeboard on air snow interface, hf, and the radar altimeter measure the freeboard on ice snow interface in the presence of dry snow and cold conditions (Figure 17).
The CryoSat retrieval algorithm has been investigated and improved after comparison with a number of other data obtained from submarine profiles, Upward-Looking Sonar (ULS) moorings, AUVs, airborne data and other. A sensitivity analysis was carried out to quantify how changes in ice density and snow depth has impact on sea ice thickness (SIT) retrievals from freeboard measurements, using the hydrostatic equilibrium equation. The results of this sensitivity analysis is shown in Fig. 18.
Validation of ice thickness retrievals from radar altimeter data from ERS and ENVISAT
Since CryoSat data has been delayed, very limited studies have been possible to carry out before the end of the project. Therefore, more work has been focussed on studying thickness retrievals from the radar altimeter data from ERS and ENVISAT. A number of examples of comparison of between thickness retrieval from radar altimeter and other thickness data are presented in the following figures
Preliminary studies of CryoSat data
A case study of CryoSat data has been conducted in the Fram Strait. Met.no has extracted relevant data from CryoSat-2 to determine how to improve its capability to detect sea ice thickness by comparing it with available in situ measurements, Synthetic Aperture Radar (SAR) data, and sea ice charts as dependent variables for ice thickness proxies. The following describes the background and application by Met.no for each data source.
Average waveforms from the level 1 data were used to determine if a criteria can be established in which radar altimetry measurements can detect sea ice thickness variations based on measurements from the waveform amplitudes. Cryosat-2 SAR mode level 1b data waveforms were converted to power in Watts with knowledge of the scale factor and power. Characteristic waveforms over sea ice show rougher signatures due to the irregularity of surface features. However, indicative patterns that include new thin ice or leads should display these features as having the highest amplitudes, whereas smaller waveforms represent surface roughness. Depending on where these open water or thin ice areas occur within the waveform, this information can theoretically be used to infer sea ice thickness. Waveforms are clearly defined to determine whether the surface is ocean or sea ice (Figure 24). Indicative features show a small peak prior to a dramatic larger peak due to noise from the reflection of surrounding elevated features. Several passes were combined and overlain on to the Radarsat-2 SAR data to illustrate how well it can detect areas of open water (Figure 25). (http://www.altimetry.info/html/use_cases/data_use_case_cryosat_2-2_en.html).
WP7 Forecasting of sea ice and icebergs
The sea ice forecasting system for the Barents and Kara Seas has been developed to produce forecasts of ice and ocean variables every week for up to 7 days. The system is based on the TOPAZ model and data assimilation system developed at NERSC (TP4), which is the main forecasting system for the Arctic region within the MyOcean project. The TOPAZ ice-ocean data and model system is the combination of the Hybrid Coordinate Ocean Model (HYCOM, Bleck 2002, and the Ensemble Kalman Filter (EnKF, Evensen, 2009) and covers the Nordic and Arctic Seas at a horizontal resolution of about 12-16 km (Bertino and Lisæter, 2008, Sakov et al. 2012, see http://topaz.nersc.no). The models use 28 hybrid z-isopycnal layers in the vertical. The TOPAZ system is today operational at the Norwegian Meteorological Institute (met.no) and daily forecast and download of data are available within the MYOCEAN platform (myocean.met.no/ARC-MFC).
The regional model over the Barents and Kara Seas (BS1) covers the area of the Barents Sea and the Kara Sea as well as a some parts of the areas in the Fram Strait, Nordic Seas, and Arctic Ocean, as shown in Fig. 27.
A high-resolution 510x450 grid is applied, that gives approximately a 5 km horizontal resolution. Nesting conditions, taken from the assimilated TOPAZ system, and tidal forcing are applied at the open boundaries. River (TRIP) and atmospheric forcing (ECMWF) are applied.
The applied sea ice model in the NERSC HYCOM model system distinguishes between the sea ice rheology in the consolidated ice pack and in the marginal ice zone (MIZ). The rheology used today in the consolidated ice pack is an elastic-viscous-plastic (EVP) model based on Hunke and Dukowicz 1997. A new MIZ rheology has been implemented into the NERSC HYCOM model system (Shen et al, 1987), where internal stresses are based on statistics of chaotic collisions between circular floes of a given size. The rheology has been further developed to include the mechanical breaking of ice floes using the mechanical parameters of sea ice (Dumont et al, 2011).
A wave-in-ice module (WIM) has been implemented into the system, where surface waves are propagated into ice-covered areas and break up the ice into smaller flows (Williams et al, 2012a, 2013a,b). The WIM model propagates wave characteristics, taken from and external wave forecast model, into the ice and keeps wave characteristics in the model memory. Surface waves impose strain in the ice and may break up the ice into smaller floes. The waves dissipate energy travelling into the ice, due to internal ice resistance and wave reflection, and will at some position be too weak to break up the ice. With the WIM model we are able to define a new criteria for the MIZ based on the floe size, here set to 200 m. In the present setup wave data is taken from the 10 km WAM North Sea forecast model (WAMNSEA10km) operated by the Meteorology Institute in Norway. A nearest point solution is applied to convert the wave characteristics onto the model grid.
The impact of the WIM module are clearly seen on the 26th of Mars 2013, when waves with large amplitude reach the ice edge, see Figure 28 below. The waves taken from an external model reach the ice edge (subplot c) and break up the ice into smaller floes (subplot d) south of Svalbard. The criterion for the MIZ rheology is set to dfloe<200m, so a new large area where maximum floe size is less than 200m are define as MIZ area, and will then apply the MIZ type of sea ice rheology.
The BS1 model runs on a daily basis producing 3 days forecast of sea ice and ocean conditions. The nesting cycle and the forecast procedure are described in Figure 29. The figure shows the weekly forecast cycle starting at a Tuesday, here described as Day0. After one week on the next Tuesday, here Day+7, the cycle repeats itself. The forecast system runs, and is dependent on forecast products on several computer servers, AT NERSC. The main server in the forecast system is HEXAGON, the supercomputer system at the University of Bergen, from where the forecast system is initiated by running shell scripts using job scheduler Crontab functions. The scripts starts to download forecast products; atmospheric forecast ECMWFR from VILJE (ntnu.no) and wave forecast WAMNNSEA10km, TP4 restart files as initial conditions, and OSI-SAF sea ice concentration from the ftp server at myocean.met.no. The outer model, TP4, is running ones a week, giving nesting condition to the inner model, BS1. The inner model runs on a daily basis using daily updated forecast products. When the BS1 forecast model is ready, forecast and validation figures are produced and send over to the NANSEN server at nersc.no. The figures are sorted and a script is initiated that update the web page at the NANSEN server. The system is dependent on four different servers and four different forecast products, see Figure 29.
A webpage is established and maintained where the daily sea ice forecast, for the next three proceeding days, see EVP+MIZ below. An automatic system is setup that downloads forcing fields, validation data, run the TP4 and the BS1 model, produce figures, and update the webpage. The webpage also include a Validation section where earlier forecasts are compared to OSI-SAF sea ice concentration fields. A parallel forecast system and webpage are set up for comparison; presenting results using the newly developed wave-in-ice module, see EVP+MIZ+WIM below. The WIM forecast is only for the next two proceeding days, due to the limitation of the + 60h surface wave forecast.
• EVP+MIZ: http://topaz.nersc.no/Knut/IceForecast/Barents
• EVP+MIZ+WIM: http://topaz.nersc.no/Knut/IceForecast/Barents2
WP8 Data integration and validation
This section describes how various SIDARUS products have been validated and used together with other data from the project.
Validation of ice-water discrimination from SAR
Sea ice classification results calculated using SVM technique have been compared with Met.no (Norwegian Meteorological Institute) ice charts from the operational ice charting service. Validation of Arctic ice products is always a challenging task due to lack of ground truth data. As a substitute, our product has been intercompared with a manual sea ice product produced by Met.no.
Met.no ice charts are produced by ice analysts at the Norwegian Sea Ice Service using the following data sources: high resolution microwave Synthetic Aperture Radar data (Radarsat), low resolution microwave SSM/I and SSMIS data (DMSP), MODIS data (Terra and Aqua) and AVHRR data from NOAA.
The procedure of Met.no and SVM classification comparison is illustrated by Fig. 30. Met.no sea ice type data has been reclassified into charts with only three classes: open water (OW), where sea ice concentration values on the original ice map were in the range from 0 to 15% , sea ice, where sea ice concentration values were in the range from 15% to 100% and land. Our results of SVM classification have been reclassified likewise - into three classes - OW, ice and land. In the comparison Met.no ice charts have been assumed to represent correct classification and we calculated the confusion matrix (error matrix) keeping in mind that assumption.
The overall accuracies of SVM classification were estimated for about 776 Radarsat-2 SAR images received in the period from March – November 2013. The average values of total accuracies for each month are presented in the table below. The accuracy is the ratio of the number of similarly classified pixels on both (Met.no and SVM) ice maps to the whole number of ice and water pixels in the image. The error is the ratio of the number of incorrectly classified pixels of each class to the total number of pixels of that class.
The average accuracy of our sea ice classification for the period March-November 2013 is 0.89 being higher in winter (0.92) and lower in summer (0.86). The accuracy is lower in summer months because the SVM classification algorithm was tuned using winter data.
Below an example of comparison of automated SAR classification and Met.no ice charts are presented. The images are shown in groups. Every group consists of two satellite images and two ice charts: Radarsat-2 HH image and Terra Modis image represent the ice conditions in the scene from the microwave and visual point of view, and two ice charts - the result of our SVM classification and the Met.no sea ice chart - demonstrate how ice conditions are described by automated classification and sea ice analysts. Additionally the confusion matrix is given for every particular pair of ice charts (SVM's result vs. Met.no). In the matrix the correctness of classification for different classes based on pixel-by-pixel differences is shown. The structure of the accuracy/error matrix is the following table:
Validation of ice drift and deformation from SAR
The validation of ice drift fields retrieved from satellite imagery can be carried out using drifting buoys. However, the number of buoys deployed in Arctic or Antarctic waters is small, and in a lot of scenes, a buoy may not be present at all. But even if this is the case, only a very limited area in the satellite-derived drift map is covered by the buoy track. Another well-suited possibility is the manual generation of a reference drift field by an experienced operator, which is then used for comparison with the automatically retrieved result. This procedure can only be carried out for a few examples but not for every image pair used in operational mapping. In the framework of the SIDARUS project, a proxy for the accuracy was developed that provides information whether the retrieved drift vectors are reliable. A robust method is to carry out the drift calculations two times, the second time using image 2 as starting point (back-matching). This approach, however, doubles the computation time. Therefore, another measure is introduced in addition that combines the analysis of textural and correlation parameters in a confidence factor (CFA). The CFA takes into account six parameters. Three of them are from the field of image texture analysis, namely the mean intensity gradient (MIG), the mean gradient slope (MGS), and the variance-to-squared-mean ratio (VMR). The fourth is simply an intensity threshold that reduces the effect of mirror reflections from the ice. These four parameters characterize the properties of each individual image but do not reflect any links between a pair of images. Therefore the correlation coefficient and its confidence interval are included in the reliability assessment. A result of such a reliability assessment is shown in Fig. 32. Sea ice deformation parameters are calculated using the drift field as input, hence their accuracy depends on the accuracy of the drift vectors. Also the ratio between spatial extension of the deformation structure and the grid used to calculate the drift, as well as the spatial resolution of the SAR images, have an influence on the uncertainties in the deformation parameters. Quantitative studies of the accuracy of different deformation parameters were not planned for SIDARUS but are recently carried out as a follow-on investigation.
Validation of the Barents – Kara Sea model
For every day the model is run, a validation is performed where the first day forecast four days earlier of sea ice concentration is compared to OSI-SAF sea ice concentrations. The 15% OSI-SAF concentration line is plotted on top of the model sea ice concentration and on top of other forecast products.
Further validation has been done off-line. In general does the forecast model follow the seasonal variability and respond correct to atmospheric forcing, though without direct assimilation within the regional model, some major deviations from measurements are seen. I) The transition from no ice to high sea ice concentration is to sharp in the model compared to OSI-SAF data, see Figure 33 a) and b). II) To much sea ice is seen north of Svalbard and west of Novaya Zemlya during most of the winter season. III) Even in sea ice thickness does the model have a too sharp transition from thin towards thicker sea ice compared to SMOS sea ice thickness, see Figure 33 c) and d).
Validation of thin ice from SMOS
Validation of thin ice is an important but complicated issue, since it is too thin to stand or walk on it. Therefore the validation of a thin ice thickness retrieval has to rely on other remote sensing data. One source of thin ice data was a night time thermal imagery sea ice thickness retrieval from the Moderate-resolution Imaging Spectroradiometer (MODIS). For Winter 2010/11 71 MODIS scenes are analyzed and compared with the SMOS retrieval showing good agreement especially in lower ice thicknesses as can be seen from Fig. 34.
Another source of validation Data was the AWI EM-bird instrument which can be carried by a helicopter or airplane which measures every few meters the ice thickness of a footprint of about 50m diameter. Fig. 35 shows one EM bird flight track and the corresponding SMOS sea ice thickness retrieval. The logarithmic histograms on the left show the distribution of EM bird measured thicknesses with their mean, median, and the SMOS retrieved values at that location number from the right-hand side. Here it is visible that the SMOS retrieval is in nearly all cases exactly at the peak occurrence of EM-bird measured sea ice thickness.
The validation shows that the sea ice thickness retrieval from SMOS works within the indicated limits.
Validation of CryoSat ice thickness
For validation of the CryoSat ice thickness data we have used a set of Radarsat-2 data in Fine Quad-Pole SLC and SGF beam mode. The satellite data have been acquired to coincide with in-situ measurements from two cruises in the Arctic: KV Svalbard (6-13 April, 2011) and Arctic Sunrise (7-12 July, 2012). Fine quad-pol Radarsat-2 SAR data was used as a ground-truth proxy for sea ice type when compared with CryoSat 2 level 1 waveforms. The CryoSat 2 SIRAL instrument is programmed to operate in SAR mode over sea ice areas, as its measurement footprint has a reduced surface area that allows the detection of smaller sea ice features. Though the use of level 2 CryoSat-2 data was preferred for this comparison due to the inclusion of multiple parameters (i.e. retracker, sea surface height, freeboard, elevation..etc.),investigation of the available products found that these fields had not been calculated. Therefore, we extracted the average waveforms from the level 1 data and converted them to power in Watts with knowledge of the scale factor and power.
Characteristic waveforms over sea ice can theoretically be used to infer sea ice thickness depending on where open water or thin ice areas occur within the waveform. However, in order for these waveforms to accurately depict the surface roughness it is necessary to implement the appropriate retracking algorithm. The following retracking algorithms are currently available:
• UCL: ESA retracker
• AWI: Threshold-Spline-Retracker Algorithm
• NOAA: Ocean height based on Maximum Likelihood Estimator
• FMI: Open water and new ice threshold retracker with a Gaussian and Gaussian + exponential fit
• Traditional OCOG retracker
• Primary peak OCOG retracker
Though several retrackers have been used with previous corrections, specific conditions require different methods of fitting the tracking point on the leading edge and the algorithms vary with each mode. The level 2 data implemented an Offset Centre-of-Gravity (OCOG) and an OCOG threshold retracker but requires further evaluation to resolve errors in the return.
However, in order for these waveforms to accurately depict the surface roughness it is necessary to implement the appropriate retracking algorithm to determine at which point the waveform is actually measuring the surface from a nadir view rather than showing effects of related to noise from how the signal varies in the range direction. Though several retrackers have been used with previous corrections, specific conditions require different methods of fitting the tracking point on the leading edge and the algorithms vary with each mode. The level 2 data implemented an OCOG and an OCOG threshold retracker but requires further evaluation to resolve errors in the return. Therefore, it will be necessary for Met.no to customize thresholds and parameters to the level 1b data to fit the needs of doing a robust comparison with SAR and in situ data.
Validation of ice albedo and meltponds
The Melt Pond Detection (MPD) algorithm has been validated using a dataset of validation data shown in Table 8.1. The data can be divided into two categories: in situ data (Fig. 36) and airborne data (Fig. 36). For each category of validation data, a collocation and comparison of retrieved and field value has been performed. For the in situ data, the correlation coefficient is 0.526. The observed scatter can be connected to low quality of some validation data (visual estimation of melt pond fraction) and different spatial resolution of the retrieval and the validation data. Here only one example of airborne validation is shown (Fig. 36), whereas the whole dataset comprises 10 comparison cases. The correspondence of the MPD retrieved values to the airborne data is reasonably good (40% field value against 50% satellite) with several exceptions which can be explained due to inability of MERIS to resolve the difference between melt pond and so called “blue ice” (sea ice without top scattering layer).
The validation of the MPD retrieval against field data is a challenging task due to lack of validation data which can be collocated with the MERIS data spatially and temporally without being sorted out due to cloud coverage. Another possibility to validate the MPD retrieval is comparison to another remote sensing product. Such comparison attempt has been performed (not shown here) and showed some discrepancies between the MPD retrieval and algorithm by Roesel et al., 2011, from MODIS data. Due to different overflight times of the two sensors (MODIS and MERIS), the cloud coverage and surface coverage might have been slightly different. On the scale of weeks (weekly averages have been compared), this could have caused some discrepancy. In addition, the difference in accounting for surface reflectance within both retrievals could have been the reason for the disagreement. Further comparison is needed to clarify the issue.
Potential Impact:
The results of SIDARUS are expected to have impact on climate research as well as marine operations and environmental protection in polar regions. The Arctic and Antarctic is dominated by ice-covered oceans and coasts. The regions are exposed to climate change with significant impact on the cryosphere and the environment that is affected by the presence of ice. In the Arctic the global warming is at roughly twice the global average rate, with a dramatic reduction in summer sea ice extent as one of the clearest indicators of this trend. Physical and biological processes are being transformed across the entire regions while climate feedback mechanisms in the Arctic’s changing atmospheric and oceanic dynamics impact at global scales. The Arctic regions offer vast areas of hydrocarbon resources that have just started to be exploited. The Arctic Ocean is surrounded by continental shelves, where in particular the huge Siberian shelf covering the eastern hemisphere, extending from the Barents Sea to the Chukchi Sea.
The ongoing changes in Arctic climate with increasing temperatures and decreasing sea ice cover have stimulated the interest for oil and gas exploration in several Arctic areas. A reduction of the sea ice area opens up the possibility to access new areas of the Arctic Ocean where hydrocarbon resources can be exploited and transported to the markets. The main Arctic areas where large-scale offshore exploration have started are: Sakhalin in Sea of Okhotsk, North Slope of Alaska, Cook Inlet, Grand Banks of Newfoundland, Barents Sea (Snøhvit field and the upcoming Shtokman field) and the Pechora Sea. A map of regions in the Arctic with potential oil and gas fields is shown in Fig. 1.
All these areas have seasonal sea ice cover and some have icebergs that put severe constraints on design and operation of installations and on transport solutions. Even if the sea ice cover is decreasing and is expected to diminish further in the coming decades, the sea ice will still remain a dominant factor in most of the exploration areas in the winter season. In the summer months, however, less sea ice will provide access to offshore areas in Canada, Greenland and on the eastern Siberian shelf that were previously inaccessible due to sea ice. Improved sea ice end iceberg monitoring and forecasting in the polar regions will have large socio-economic impact due to climate change combined with increased human activities. The global demand for energy has stimulted the political interest for the Arctic Ocean and several countries have started investigations of the continental shelves. Under the UN Convention on the Law of the Sea, a country can claim exclusive economic rights within 200 miles (Fig.2) If a country can prove that its continental shelf extends beyond the 200-mile economic zone, it can claim similar rights over a larger area. All States involved in the Arctic Ocean continental shelf have ratified the Convention except the USA.
Ice information products, such as charts and forecasts, are provided by the national meteorological or oceanographic services of countries with activities in ice-affected waters. Currently, shipping is the primary user of these products (Fig. 3). They are created by combining data from satellites, in situ sensors, and aerial and shipboard observations. Each source has strengths and weaknesses. In situ sensors, aerial surveys and ships provide specific but sparse information, and aerial surveys are expensive. Satellite data are not as detailed, but they are systematic, cost-effective and cover wide areas. A variety of satellite sensors provide data at varying resolutions, spatial scales and costs. Because ice can be highly dynamic, ice products must be synthesised quickly (in 1-6 hours) and regularly (every 6-24 hours, every day). Synthesis of data archives allows for statistical analysis and prediction.
The results of SIDARUS will improve sea ice information and be useful for offshore industry and shipping polar areas, scientists working in climate and environmental research and organisations working with protection of sea ice habitats. The existing GMES provide services large-scale sea ice concentration and ice drift maps from satellite data and ice-ocean forecasting for the whole Arctic with about 20 km grid cells. The forecasting products are provided by the TOPAZ system at NERSC (http://topaz.nersc.no). Large scale sea ice charts are provided by University of Bremen (http://iup.physik.uni-bremen.de:8084/amsr/amsre.html) the ice drift maps are produced by Ifremer (http://cersat.ifremer.fr/data/discovery) and regional high-resolution ice charts for the Svalbard area by Met Norway. Other products cover Greenland waters, Canadian areas and Antarctica.
SIDARUS will provide local ice information for areas around planned drilling areas and along sailing routes, using high resolution SAR and optical satellite images, with pixel size ≈ 10 m or smaller. The products from the satellite images will include maps of ice concentration, ice drift and ice type classification, the latter including a separating of smooth level and deformed ice (ridges, shear zones), open water, leads, and areas of thin ice. Satellite data alone cannot provide all required information about sea ice and icebergs. Therefore, SIDARUS will also include additional data, in particular ice thickness using airborne surveys with the HEM system and in situ observations with the EM31 instrument and with multibeam sonar under the ice. SIDARUS will also provide sea ice and iceberg forecasting for the Barents Sea, using high-resolution models nested to the TOPAZ system, and a web-based ice information system where new as well as archived data and model simulations are made available. This will enable users to obtain both statistical information based on previous observations and model simulations, new observations based on multisensory approach, and short-term (up to 10 days) forecasting using NCEP forcing fields. The services developed in SIDARUS will first be implemented for the Barents Sea region, with focus on the Shtokman area that will start to be developed in 2012. The services will later be implemented in other areas where offshore industry plan to start operations. In the Antarctic, iceberg monitoring and forecasting will be implemented. The services will complement and extend the present GMES Marine Core services.
There is a growing interest for combined information on sea ice mapping and animal tracking provided, for example for polar bears, as expressed by Dr Andrew Rocher, Professor at University of Alberta, and Chair of the IUCN/SSC Polar Bear Specialist Group: “The remoteness of polar bear habitat, the low density of bears, and the long periods of winter darkness preclude most field research methods. Satellite telemetry has thus played a major role in developing insights into their ecology and aiding their management, dating back to the 1980s”. High resolution products provided by SIDARUS would also help the wildlife user group: Earth Observation data at medium resolution (12.5 km) such as AMSR-E passive microwave instrument to know the actual sea ice situation. For example, the Alaska Science Centre of the US Geological Survey tracked walruses in Summer 2007 and found them in open water in end of July. They presumed that there were remnant ice floes that the walruses were using as haulout platforms. However this high resolution information was not available from the medium resolution products. Examples of user organisations working with protection of sea ice habitats are: (1) The Institute of Evolution and Ecology Problems (Russian Academy of Sciences). They are already using satellite data (Argos) for wildlife surveillance and they are keen on this new concept involving ice data. They are very interested in ice density and the breathing facility of mammals under the ice. This would be a totally new way of understanding wildlife behavior. (2) The Alberta University, Canada. The Department of Biological Sciences has completed long term studies on Ecology, conservation, and management of large Arctic mammals focusing on polar bears. Interests centre on limiting and regulating factors of polar bear populations including habitat use, harvest effects, and predator-prey relationships. Current research includes assessment of the effects of climate change on polar bears. That University has already worked on joint projects with the Wegner Institute in Germany. (3) The Greenland Institute of Natural Resources, Denmark. The Greenland Institute is ready to share its humpty whales tracking data in Iceland. The Institute plans to study narwhales In East Greenland during the summer 2010. These mammals will presumably move around the sea ice in Greenland and their movements will be correlated with ice information from SIDARUS. Also, the Institute is tagging polar bears with Argos collars between Canada and Greenland.
The Intergovernmental Panel on Climate Change (IPCC) has provided several scenarios for future climate change in different parts of the world [5]. For the Arctic, the observed trend of reduced ice extent will continue and the summer ice extent may disappear towards the end of the century. The record low ice extent in September 2007 and 2012 suggested that the summer ice may disappear much sooner (Fig.4). In the early 1980s the summer ice area was about 6.5 mill km2, and it diminished by about 8 % per decade until 2006. In 2007 it the area shrunk to less than 4.5 mill km2. The reduction of ice area of more than 2 mill km2 corresponds to the size of Greenland. When the summer ice disappears, the thick multiyear ice will be absent, leaving the Arctic Ocean with a thinner ice cover in the winter season. The disappearance of multiyear ice combined with longer melt season will have important operational implications, leading to greater access and longer navigation season for shipping around the Arctic basin.
The IPCC scenarios suggest that there will be a substantial warming in Arctic and sub-Arctic areas compared to the present situation, which is closely linked to the reduction of the sea ice. The surface air temperature in many climate model projections shows a 6-8°C warming over the ocean during winter, with a less dramatic change in terrestrial regions. With higher temperature and reduction of the ice cover, the marginal ice zone will move poleward, leaving the coastal and shelf areas ice-free in the summer.
With a smaller area covered by sea ice, more heat from solar radiation will be absorbed in the ocean, leading to increased ocean temperature. A warmer ocean will in turn reduce the amount of sea ice formed in the following winter. This is a so-called positive feedback mechanism, leading to enhanced warming in the Arctic. Another effect of a warmer Arctic is more clouds and precipitation. Increased fog will result sin in poorer surface visibility, which is an obstacle for many operations. More frequent and stronger storms can also be expected in the sub-Arctic areas. Vessel icing could also increase in these areas, especially during outbreak of cold Arctic continental air masses.
How can we assess the climate change impact in the various sub-Arctic regions where offshore operations are foreseen ? During winter, the central Arctic and all peripheral seas including the Greenland Sea, Bering Sea, and Gulf of St. Lawrence will continue to have significant ice cover. Ice extent and thickness will generally be reduced. The Sea of Okhotsk and Sea of Japan will be ice-free for the entire year. In late summer, the entire Russian coast will be ice free, allowing navigation through the Barents, Kara, Laptev and East Siberian Seas along the entire Northern Sea Route [6]. This situation has already been observed in the last couple of summers. The Northwest Passage through the Canadian Archipelago and along the coast of Alaska will in general be ice free and navigable in summer by non-icebreaking ships. Ice will be present all year along the eastern and northern coasts of Greenland. Ice will also remain throughout the summer within and adjacent to the northern Canadian Archipelago. However, severe winters with more ice than average may also be expected due to the natural variability of the climate system. The effect of more wind and waves in ice-covered areas will be increased ridging and stamukhas in near coastal regions. The iceberg situation in different parts of the Arctic is difficult to assess, but it is likely that more icebergs can occur in some years as a consequence of diminishing Arctic glaciers. Arctic shipping is expected to increase as a consequence of less sea ice and more offshore exploration.
The possible consequences of increased oil and gas exploration in the vulnerable Arctic environment is a controversial issue. The Arctic ecosystems are already today exposed to severe treats due to the effects of a warmer climate. The climate effect comes in addition to the latent risk of radioactive contamination due to extensive nuclear bomb testing in the Russian Arctic in the previous decades. The storage of nuclear waste from scrapped reactors is also a severe risk factor, because it is not clear how safe this storage will be in the future. A growing oil and gas industry operation on land as well as at sea will increase the pressure on the environment with increased risk of accidents that can have severe and long-lasting negative effects on ecosystems. A worst-case scenario is an Exxon-Valdez type of accident that occurred in Alaska in 1989. The ecosystems in the area affected by this accident are still marked by this oil pollution disaster, almost 20 years after it happened [8]. The environmental impact of oil and gas exploration will be higher in the Arctic compared to other areas in the world. This calls for new technologies to ensure safe operations as well as legislative norms that regulate the activities. These factors are not in place yet and need to be developed.
In conclusion, offshore operations in the Arctic will be more feasible as a consequence of the climate change, leading to less sea ice and warmer temperatures, The costs of operations, however, will be high due to extreme ice and weather conditions and requirements to operate with minimum risks to harm the vulnerable Arctic environment. This requires that adequate ice monitoring and forecasting systems are developed, validated and implemented. SIDARUS results can make a significant contribution to this goal.
List of Websites:
http://sidarus.nersc.no
Prof. Stein Sandven
Director
Nansen Environmental and Remote Sensing Center (NERSC)
Tel: +47 55 20 58 00
Fax: +47 55 20 58 01
E-mail: stein.sandven@nersc.no
