Enhancing Arctic Weather Forecasting through Integrated Use of Infrared Satellite Observations over Sea Ice
Project concept
AWFInfra addresses a key weakness in Arctic weather forecasting: uncertainty in sea-ice surface temperature estimates caused by sparse in situ observations, cloud contamination, mixed surfaces, and strong local gradients. The project improves both physical and machine-learning forecasting systems by integrating high-resolution infrared satellite observations with in situ measurements and model-based analyses.
The project emphasizes use of Sentinel-3 SLSTR, MODIS, and VIIRS products, and evaluates how observational biases propagate into AROME-Arctic analyses, CARRA reanalyses, and ML forecasting workflows. It also develops observation-aware ML integration in the Anemoi framework to reduce dependence on biased model-only inputs.
Scientific objectives
- Advance operational and scientific use of infrared satellite surface temperature retrievals over sea ice.
- Quantify and correct systematic biases in satellite-derived surface temperature datasets.
- Assess impacts of observational biases on NWP analyses, reanalyses, and ML-based forecasts.
- Develop and test direct satellite-to-ML integration methods in Arctic forecasting applications.
- Support next-generation European Arctic forecasting developments through coordinated intercomparison activities.
Methodological foundation
AWFInfra combines satellite observations with distributed in situ infrared radiometer measurements from recent Arctic campaigns, enabling robust bias characterization and uncertainty quantification. This cross-comparison framework provides physically grounded correction strategies and creates a pathway toward observation-aware ML forecasting.
Implementation and outcomes
The project is structured around management and coordination, bias characterization against in situ and model products, and ML integration of satellite observations. Key outcomes include bias-assessed and quality-controlled datasets, tested observation-aware ML prototype capability, and peer-reviewed scientific deliverables on bias diagnosis and ML integration in Arctic prediction systems.
Funding acknowledgement
This project is supported by the Norwegian Space Centre.
