The Earth Observation Laboratory is a research unit within the Department of Physics and Technology at UIT The Arctic University of Norway. We conduct research in the field of satellite remote sensing applied to earth observation. Our main focus is on analysis of synthetic aperture radar (SAR) images and development of image analysis tools based on polarimetry, advanced statistics, pattern recognition and signal processing." /> The Earth Observation Laboratory is a research unit within the Department of Physics and Technology at UIT The Arctic University of Norway. We conduct research in the field of satellite remote sensing applied to earth observation. Our main focus is on analysis of synthetic aperture radar (SAR) images and development of image analysis tools based on polarimetry, advanced statistics, pattern recognition and signal processing." />

Master Thesis Proposals

 

Presented here are thesis proposals for students, supervised by staff at the Earth Observation group.  

The scope of these proposals can be adjusted and are equally suitable for students from the 5-year integrated masters (30sp) and the 2-year master’s program in Physics (60sp) 

 


 

The Earth Observation group currently have the following project proposals:

 

 


 

 

Would like to learn about the latest technological solutions for monitoring various Arctic marine environments?

Do you want to develop machine learning methods, which integrate the physical properties of ocean remote sensing? Do you want to learn how is it possible to retrieve information about microscopic marine organisms by acquiring data 800 km above sea surface? Do you want to obtain skills highly relevant for the industry and academia?If your answer is yes, I am looking forward hearing from you.

Contact: Katalin Blix (katalin.blix@uit.no)

To the top

 

Characterization of low backscatter areas in SAR imagery using Passive Microwave and thermal sensors

Synthetic aperture radar (SAR) images are operationally used for sea ice type characterization and ice edge monitoring in the Arctic. The marginal ice zone (MIZ) is challenging for sea ice classification and is experiencing highly varying wind conditions. The varying wind conditions results in a wide range of SAR backscatter values. High wind speeds make the backscatter values similar to the thicker sea ice and the low wind reduce the backscatter values. At the same time is the MIZ an area with a high degree of new ice formation, also resulting in low backscatter values. Due to the low SAR signal within these areas it may not be possible to correctly classify these areas using the SAR information alone. However, inclusion of passive microwave and thermal band from optical sensors can aid this.

The goal within this project is to detect the low backscatter areas and then accurately classify them. The project will use a low backscatter segmentation algorithm, developed within the group, to identify the low backscatter areas using Sentinel-1 images. Sentinel-1 provides daily images over the European Arctic and is therefore ideal to use here as changes detection between two images can further aid the classification procedure.

The Passive microwave and optical sensors will be used to assess if they can allow for characterization of these low backscatter areas. It is expected that areas with too high surface temperatures could be identified using the thermal information and the sensitivity to thinner sea ice types within the passive microwave can identify the thinner sea ice areas. The different resolution for the different satellite sensors needs to be addressed, e.g. the passive microwave sensors has comparatively large pixel resolution compared to Sentinel-1 and will therefore not be able to aid if the dark features are too narrow or too close to the land. The Barents Sea has a high degree of new ice formation and will be the study area for this project.

Contact: Malin Johansson (malin.johansson@uit.no)

To the top

 

Machine Learning for Sea Ice Classification

Machine learning methods have been frequently used for remote sensing applications. Lately, a new approach has been successfully introduced for estimating geophysical sea ice information from dual-polarimetric SAR data, information which in general require quad-polarimetric observations. The dual-polarimetric SAR data is from the freely available wide swath data from the Sentinel-1 satellites, which are often preferred for sea ice monitoring. This project will further study machine learning models for sea ice monitoring by using the established methodology together with machine learning classifiers to address sea ice classification. The study will in particular address the problem of distinguishing between sea ice and open water under various weather conditions, and potentially develop a tool/methodology for robust large-scale localization of the ice edge.

You will learn to process SAR data, develop and apply machine learning methods and understand sea ice remote sensing.

Contact: Katalin Blix (katalin.blix@uit.no), Martine M. Espeseth (martine.espeseth@uit.no), Torbjørn Eltoft (torbjorn.eltoft@uit.no)

To the top

 

Machine Learning in Synthetic Aperture Radar Imagery

Synthetic Aperture Radar (SAR) is the preferred imagery for several applications, since it can monitor the surface independently on the light and weather condition. The freely available wide swath data collected by SAR sensors onboard the Sentinel-1 satellites are often preferred for monitoring purposes. However, these sensors operate in dual-polarimetric (dual-pol) mode, which limits the information retrieval capacity. The project would attempt to overcome this limitation by introducing machine learning approaches, that will learn the relationship between different polarizations. This would increase the information that can be retrieved from dual-pol systems, and hence allow a broader application area. You will learn how machine learning techniques can be used to increase information retrieval capacity, process SAR data and use it in Earth Observation, and understand sea ice remote sensing.

Contact: Katalin Blix (katalin.blix@uit.no), Martine M. Espeseth (martine.espeseth@uit.no), Torbjørn Eltoft (torbjorn.eltoft@uit.no)

To the top

 

Multi-frequency SAR study of thinner sea ice and ridged areas

The Arctic sea ice landscape is rapidly transforming and this includes a thinning of the sea ice and a reduction of the sea ice extent. The dark and often cloudy environment means that satellite borne radar sensors, with multi-polarization capabilities, play a key role in the sea ice monitoring. The marginal ice zone in particular is subjected to cloud cover and fog and is also an area of specific interest for guiding ship traffic and for the maritime industry. Ice regimes differ in composition and areal fractions of ice types, ice concentration, and degree of deformation.

C-band SAR are used operationally to map presence of sea ice as well as classify different sea ice types. However, L-band SAR has been shown to be sensitive to deformed sea ice areas and in addition does the good signal-to-noise ratio at L-band enable good separation of young ice types from surrounding thicker sea ice and the open water. The larger penetration depth in L-band SAR is of general benefit for characterizing and classifying sea ice, as it improves discrimination between first (FYI) and multiyear ice (MYI), and is of specific benefits for detection and characterization of leads and thin ice areas. The low noise floor for ALOS-2 enables reliable separation of thin ice areas from thick sea ice also in the presence of frost flowers.

This project will build upon overlapping fully-polarimetric L- and C-band data collected during the MOSAiC campaign in 2019-2020 and during the N-ICE2015 campaign in 2015. The aim is to investigate the importance of the pixel resolution for the detection of the thin ice and ridges and how this varies with the two different frequencies. This can be done using both spatially and temporally overlapping medium-resolution images such as those from the Sentinel-1 or by simulating images with a lower resolution. The outcome is expected to provide answers to the size of thin ice areas and ridges can be captured in the high-resolution fully polarimetric images and how does this translate to the daily medium-resolution Sentinel-1 images.

Contact: Malin Johansson (malin.johansson@uit.no)

To the top

 

Can cold snow interefere with radar signal?

It is normally assumed that the cold snow layer is transparent for the light in microwave part of the spectrum. The actively sent pulse of light should scatter predominantly on the sea ice surface or shortly after penetrating it. Recent research shows that that does not hold for salty and wet snow (Nandan et al, 2017). This knowledge is essential for correct retrieval of sea ice thickness and snow from radar altimeters mounted on satellites (e.g. CryoSat-2, Sentinel-6 and future ESA missions).At MOSAiC (https://mosaic-expedition.org/) several thousand snow depth measurements were collected each week along several-kilometer-long transects (Itkin et al, 2020). At the same location sea ice thickness was recorded. The X-band ice radar mounted on the research ship (Oikkonen et al, 2017) was recording continuously over large area surrounding those measurements. The first look at the imagery suggests that large artificial snow accumulations can change X-band radar intensities. While the Arctic snow cover is typically thin (10-30 cm), large amounts of snow can accumulate also in naturally-occurring features of pack ice -in the pressure ridges (Liston et al, 2018). When the ridges are first formed the snow cover there is same thin as everywhere else and even blocks of bare ice are exposed. Over the time, ridges as rough features in windy environment accumulate deep snow drifts, in places over 1m deep. For this master thesis topic, the candidate will explore the development of ship radar back-scatter intensities and compare them to the snow depth observed in-situ on the ice. The candidate should be interested in sea ice geophysical processes and statistical analysis of the data. The analysis could be potentially extended to satellite radar data and radar signal modeling.

Literature:

- Nandan, V., Geldsetzer, T., Yackel, J., Mahmud, M., Scharien, R., Howell, S., ... Else, B. (2017). Effect of snow salinity on CryoSat‐2 Arctic first‐year sea ice freeboard measurements. Geophysical Research Letters, 44, 10,419–10,426. https://doi.org/10.1002/2017GL074506

- Oikkonen, A., Haapala, J., Lensu, M., Karvonen, J., andItkin, P. (2017), Small‐scale sea ice deformation during N‐ICE2015: From compact pack ice to marginal ice zone, J. Geophys. Res. Oceans, 122, 5105–5120, doi:10.1002/2016JC012387.

- Liston, G. E., Polashenski, C., Rösel, A., Itkin, P., King, J., Merkouriadi, I., & Haapala, J. (2018). A distributed snow‐evolution model for sea‐ice applications (SnowModel). Journal of Geophysical Research: Oceans, 123, 3786‐ 3810. https://doi.org/10.1002/2017JC013706

- Itkin et al 2020: Winter sea ice and snow mass balance from repeated transects in MOSAiC Central Observatory, AGU Fall Meeting, December 2020, online presentation available at: https://agu.confex.com/agu/fm20/videogateway.cgi/id/22222?recordingid=22222

Contact: Polona (polona.itken@uit.no), Anthony(anthony.p.doulgeris@uit.no)

To the top

 

Influence of height, distance, shape and orientation of pressure ridges on the X-band ice ship radar

The images from navigational sea ice ship radar could be used to develop sea ice classification (e.g. see pioneering work with ship radar by Kaspersen (2017), and sea ice classification examples in Lohse et al, 2020). The images with radar intensities can be comparedto airborne data: Airborne Laser Scanner (ALS) from helicopter on MOSAiC expedition (for example data from previous campaign see: Itkin et al, 2018) or more spatially limited digital elevation model (DEM) from drone at the Nansen Legacy. Some data is available as ‘raw’ radar back-scatter, but majority of the data is already converted to signal intensity. How can we best use such limited data? Additional information that can be used is in-situ data from ice cores and snow pits inside the radar footprint, ice observations from the ship, and back trajectory analysis of SAR images and passive microwave products. The candidate should be interested in image analysis, sea ice geophysical processes and statistical analysis of the data. The topic has a potential todevelop into a PhD thesis.

Literature:

- Kaspersen K. M., 2017, Master Thesis at UiT: https://munin.uit.no/handle/10037/11627

- Lohse J, Doulgeris AP,Dierking W (2020). Mapping sea-ice types fromSentinel-1 considering the surface-typedependent effect of incidence angle. Annals ofGlaciology1–11. https://doi.org/10.1017/aog.2020.45

- Itkin, P., Spreen, G., Hvidegaard, S. M., Skourup, H., Wilkinson, J., Gerland, S., & Granskog, M. A. (2018). Contribution of deformation to sea ice mass balance: A case study from an N‐ICE2015 storm. Geophysical Research Letters, 45, 789–796. https://doi.org/10.1002/2017GL076056

- Cruise reports from MOSAiC expedition (in preparation, drafts available as internal material on request from polona.itkin@uit.no)•Cruise report from Nansen Legacy cruise SSQ2 –April/May 2021 (draft available in June 2021)

Contact: Polona (polona.itken@uit.no), Anthony(anthony.p.doulgeris@uit.no)

To the top

 

Efficient interpolation of satellite and numerical data

A challenge with using satellite data together with numerical model data or other satellitedata is that each product has a different sampling grid. Synthetic aperture radar and passivemicrowave imagery is often provided in the sensor geometry (swath geometry). Optical satellitedata is often provided on a map projected grid. Numerical models may exist on a regularprojected grid or even an irregular grid. Hence, efficient joint analysis of such data sourcesrequires resampling (interpolation) between the different data domains.

Satellite images also tend to be large (often several GB per image). This means that in additionto being accurate, the resampling operations must also be computationally efficient to allowfast generation of results. This project would therefore focus on implementation of a set ofparallelized routines for interpolating different data sources onto the same grid.

The project is very flexible with respect to the application, and the student could for exampleapply the resampling to:

•Multi-resolution analyses with SAR and passive microwave

•Joint analysis of SAR data with numerical data such as surface temperature, wind fields,etc.

•Co-registration of pairs from Interferometric SAR images

•On-demand generation of animated image timeseries similar to the Sentinel Hub EOBrowser (https://www.sentinel-hub.com/explore/eobrowser/)

Contact persons: Thomas (thomas.kramer@uit.no), Anthony (anthony.p.doulgeris@uit.no)

To the top

 

New approach to SAR Terrain correction

Synthetic aperture radar (SAR) back-scattered signals are dependent on both the surface prop-erties (making them useful for remote sensing) and the viewing geometry (which complicatesthe interpretation). The signal strength varies systematically with the incidence angle withan approximately exponential relation, or linear in the log-domain (in decibels, dB). This istraditionally corrected with a single rate for the entire image as a pre-processing step known asincidence angle correction. However, many research papers observe different rates of decreasefor different surface classes, mostly related to surface roughness properties, and so this one ratecorrection is not ideal.

We have recently had great success by modelling thisincidence angle effectas a class prop-erty (rather than an image property), thus allowing different rates per class. This has beendemonstrated for maritime monitoring where the satellite viewing angle is changing smoothlyacross the imaging swath on the flat ocean surface. We believe that the per-class rates mayalso help land-based classification where terrain slopes vary the signals, but this has not yetbeen demonstrated.

The project shall explore how to introduce per class land-based terrain correction into seg-mentation or classification algorithms. The first stage may search for large flat terrains, suchas large scale agriculture, to confirm its benefits in a similar case to the flat ocean case, andsubsequent cases will seek to introduce the local terrain angles (derived from Digital ElevationModels, DEMs) with different degrees of rough surface topography (from hills and glaciers tomountains) for evaluation of the approach.

Contact: Anthony (anthony.p.doulgeris@uit.no)

To the top

 

Understanding sea ice roughness and surface state from a combination of CryoSat-2 radar altimetry, ICESat-2 laser altimetry and SAR

Satellite altimeters like ESA’s CryoSat-2 and NASA’s ICESat-2 offer new perspectives for understanding sea ice properties in the Arctic. Over the past two decades sea ice in the Arctic basin has transitioned from mainly thick, old,and rough multi-year ice to thinner, smoother,and younger first-year ice types, as the northern polar climate has warmed at a rate 2-3 timesfaster than the global average.

It is critical we develop a strong understanding of the physical changes in the sea ice accompanying this decadal shift, so that we can better simulate sea ice floes in numerical models of Arctic climate. Principal challenges of numerical sea ice modelling areto predict how ice floes react to winds and ocean currents,and how quickly the ice melts from above in summer months, both of which depend closely on the sea ice surfaceroughness. However, there is not yet an establishedpan-Arctic sea ice roughness data product available from satelliteremote sensing.

Both altimetry and SAR imaging sensors are sensitive to the sea ice roughness. The plot above shows roughnessheight observations from the CryoSat-2radar altimeter overlaid on a Sentinel-1 SAR scene from the Canadian Arctic Archipelago in spring 2018. Sea ice roughness is evidently higher over older multi-year sea ice floes. This project will integrateroughness observations from anew CryoSat-2 data product https://data.bas.ac.uk/full-record.php?id=GB/NERC/BAS/PDC/01257 with those from the ICESat-2 laser altimeter ATL10 product https://nsidc.org/data/atl10. The aim will be to test the sensitivity of radar and laser altimeters to the sea ice type, surface state and roughness over a set of coincidingSAR scenes. For example, the EO groupwill be acquiring fully polarimetric RADARSAT-2 imagescoinciding with both altimeters at Cryo2Ice tracks https://cryo2ice.org/.This workwill contribute towards establishing a baseline sea ice roughnessdata product for the Arctic Ocean.

Contact: Jack Landy (jack.c.landy@uit.no)

To the top