Sea Ice Retrievals and data Assimilation in NOrway

In the High Arctic, the sea-ice cover changes and moves all the time because of the changes in temperature and wind. MET uses images from many satellites to tell where the sea ice is every day. We also use computer models to forecast where it will be in the next couple of days (like the weather forecast on Yr, but for sea ice). Because it is more difficult to predict sea ice than wind or rain, we started a new research project. The name of the project is SIRANO, which stands for Sea Ice Retrieval and data Assimilation in NOrway. SIRANO ran from 2020 to 2024. In SIRANO, we focused on the region around Svalbard and the Barents Sea. The project had three research tasks: 1) we wanted to use the available satellites each on their own; 2) we wanted to use the satellites better together; and 3) we wanted to use the satellite data to improve MET’s sea-ice forecasts in a smarter way, called data assimilation. SIRANO collaborated with the CIRFA SFI (led from University in Tromsø). CIRFA had developed automatic methods to analyze detailed satellite images called Synthetic Aperture Radar (SAR) images. SAR instruments see many details in the sea-ice cover, but images are hard to fully decipher by a computer algorithm, and there are long gaps between the images. In SIRANO, we further evaluated some of the methods developed in CIRFA. We chose to work further on a method using Machine Learning (ML). Machine Learning is a new way for computers to learn how to do tasks by looking at examples. In SIRANO, we used a special computer program called a U-Net to teach the computer to recognize ice from water in satellite images. We also assessed if slightly degrading the resolution of the SAR images improved the accuracy of the sea-ice results. Also, we worked with another type of satellite sensor called Passive Microwave Radiometers (PMR). PMRs do not show as many details as SAR, but they are easier for computers to understand and provide many images each day. In SIRANO, we created a system to process satellite data from the Japanese AMSR2 mission. This software was installed at MET Norway and has been preparing new sea ice maps for the Barents Sea region every day. Finally, towards the end of the project, we extended the U-Net technique to use both the SAR and PMR images at the same time. Using the two types of satellite imagery together, with the Machine Learning technique, gave the best results. The third task dealt with improving the assimilation of satellite data into MET's ocean and sea-ice forecast systems. Before our research in SIRANO, satellite data were collected and averaged into 12-hour maps before being used in the forecast model. During SIRANO we demonstrated that it is better to assimilate the new satellite images as they arrive instead of waiting to prepare the 12-hour maps. This gives better forecasts. The research in SIRANO was done at MET in Oslo and UiT in Tromsø, and we also worked with other researchers from around the world. In 2023, SIRANO invited the international community in Oslo for the 11th International Ice Charting Working Group Data Assimilation Workshop (IICWG-DA-11). The workshop attracted almost 100 persons (+ 40-50 online). All presentations and some posters are available online (https://iicwg-da-11.met.no/). This was a sign that the research in SIRANO was relevant. SIRANO recruited and trained a PhD-student and a PostDoc researcher who both continued their work in Earth Observation and use of satellite data. In conclusion, SIRANO was a successful research project that consolidated the collaboration between MET and UiT for the use of satellite observations in the Arctic, and led to better sea-ice observation and forecast in the Barents Sea and Svalbard regions.

SIRANO tackled some of the knowledge gaps that were hindering the full exploitation of key satellite technologies, as a support to socio-economic activities and environment monitoring in the Arctic and High North. By placing SIRANO closely with the teams that develop the forecasting services of MET, we wanted the project to contribute to these 24/7 services. This outcome is to a large extent realized at the end of SIRANO as MET’s operational ocean and ice forecasting system for the Barents Sea region was assimilating the new satellite products developed in SIRANO. Thus, we can say that the research from SIRANO had a direct outcome on the 24/7 ocean and sea ice forecasts issued by MET for the Barents Sea and Svalbard region. Not all of the satellite products developed during SIRANO reached enough maturity at the end of the project period (late 2024). For example, the joint sea-ice retrieval product based on both Synthetic Aperture Radar (SAR) and Passive Microwave Radiometer (PMR) using Machine Learning techniques were in the process of being implemented, to be monitored and possibly further developed before they could reach the operational stage (delayed outcome). One of the methodological impacts of SIRANO was to confirm the adequacy of Machine Learning techniques to jointly exploit SAR and PMR data for sea-ice applications. At the time when SIRANO started, other methodologies such as Bayesian classification were more prominent. SIRANO played a part in this transition at MET. Another future impact of SIRANO is that MET is now better prepared to take in use future satellite missions, in particular those in the Copernicus programme (like the Sentinel-1 Next Generation satellites, and the Copernicus Imaging Microwave Radiometer (CIMR). Both missions will fly in the early 2030s.

The Sea Ice Retrievals and data Assimilation in NOrway (SIRANO) project is a research collaboration between the Norwegian Meteorological Institute (MET) and UiT - the Arctic University of Norway. In the Arctic, around Svalbard and in the Barents Sea, sea-ice moves every day, because the temperature and wind change. We use pictures from many satellites to tell where the sea-ice is every day. We also use computer models to tell where sea-ice will be in the next couple of days (like the weather forecast on Yr, but for sea-ice). But it is more difficult to predict sea-ice than wind and rain, and we need to do some research to make better sea-ice forecasts. In our project, we will perform this new research. Three things that must be improved are: 1) to use satellites better individually, each on their own; 2) to use the satellites better all together; and 3) to use the satellites better together with the forecast models. The main difficulty is to combine pictures from several very different satellites. For example, a type of satellite we want to use (Synthetic Apperture Radars) sees many details, but the pictures are hard to understand by a computer algorithm, and there are data gaps. Another type of satellites (MicroWave Radiometers) does not see the small details, but can easily be interpreted by computer algorithms and takes many images a day. How can we build a third satellite picture, that will combine the pictures from these two satellites well? This is one of the questions we must answer. When we are done with our research, Norway will have new methods to use satellites and computer models to tell you where the sea-ice is, and where it is moving to. This is very useful for ships sailing or fishing around Svalbard, and for the Search and Rescue teams if there is an accident. It will also improve the weather forecasts we prepare for Yr.