Using AI methods in Global Navigation Satellite System Reflectometry (GNSS-R) remote sensing data to overcome the limitations of conventional methods in forecasting severe climate events and trends.
Today’s Helmholtz AI project showcase investigates whether deep learning can be a viable and effective method for GNSS reflectometry. The work is being carried out by researchers from the Helmholtz Centre Potsdam - German Research Centre for Geosciences (GFZ) and the German Aerospace Center (DLR).
Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?
I am Milad Asgarimehr, a postdoctoral researcher at the German Research Centre for Geosciences (GFZ) in Potsdam. I work on remote sensing, specifically techniques exploiting existing signals of positioning systems such as GPS, and study the Earth system and climate. The other researcher on this project is Lichao Mou, the Head of Visual Learning and Reasoning team at the Department “EO Data Science”, Remote Sensing Technology Institute (IMF), German Aerospace Center (DLR) and a Guest Professor at the International AI Future Lab AI4EO, Technical University of Munich (TUM). He is interested in algorithms for Earth observation data analysis and visual learning and reasoning tasks. Our project is titled “Artificial Intelligence for GNSS Reflectometry: Novel Remote Sensing of Ocean and Atmosphere (AI4GNSSR)”.
In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?
The GPS signals, and those of other positioning and navigation systems such as Galileo, GLONASS, and BeiDou, reach the Earth’s surface and are reflected. The reflected signals intercepted by receivers onboard small satellites and CubeSats can be used to extract a variety of Earth’s surface and atmosphere properties. This novel remote sensing technique is called Global Navigation Satellite System Reflectometry (GNSS-R). The low development costs ease the launch of multi-satellite GNSS-R constellations and, therefore, a remote sensing dataset with an unprecedented sampling rate. With the availability of large datasets, deep learning has become a viable and effective method in GNSS-R.
The project AI for GNSS-R (AI4GNSSR) aims at implementing deep learning for novel remote sensing data products based on spaceborne GNSS-R measurements. These include high-quality ocean surface wind speed data, especially at extreme conditions and hurricanes, and potentially precipitation over calm oceans for the first time using GNSS signals. Due to the complicated interactions of environmental and technical parameters controlling the GNSS-R measurements, methods beyond the traditional approaches are required. The AI methods can potentially overcome the limitations of the traditional methods and the imperfect theoretical knowledge in young remote sensing domains. The data products of this study will close the data gaps, especially needed to forecast severe events and identify climate trends. Besides, the knowledge derived from explainable AI will enhance the physical knowledge in this domain.
How important has the Helmholtz AI funding and platform been to carry out this project?
An inter-domain synergy is required to gather experts for different aspects of the project. Deep learning is well established in optical remote sensing and can be used to unlock the novel potentials of GNSS-R. This Helmholtz AI project brings together experts from both domains for a synergistic study and realisation of the objectives. The funding has brought new team members to GFZ and DLR, e.g. Tianqi Xiao and Daixin Zhao.