Optimization of forest radiative transfer modelling for accurate monitoring of forest traits from Earth observation time series: INFORM+

• Applicant: Dr. Katja Berger
• Subject Area: Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 563771942

The demand for spatially explicit knowledge about forest properties is continually growing to meet increasing requirements for sustainable forest management under climate change. Frequently updated understanding of structural and biochemical forest variables in mid-European landscapes is crucial for monitoring forest growth, vitality, resilience, biodiversity, and hazard susceptibility also within the context of recent European environmental policies. However, readily available forest geospatial layers are currently limited to tree cover density or forest types. The main research objective of INFORM+ is to provide a hybrid (i.e., physically based plus data-driven) and hence generic mapping of forest traits from currently available Earth observation (EO) time series, including multi- and hyperspectral scanners and Light Detection And Ranging (LiDAR). Our main research hypotheses are: (i) INFORM+, building on the existing INFORM, can simulate realistic LiDAR full-waveform and multi-/hyperspectral signatures comparable to measurements, thus enabling an improved performance in retrieving forest traits (in inverse mode) compared to the original model; (ii) spatio-temporal regularization will significantly improve forest traits retrieval accuracy compared to a non-regularized approach; (iii) hybrid retrieval strategies combining artificial neural nets and RTMs as well as prior information will provide more accurate estimations of forest traits compared to data-driven methods; and (iv) deep learning methods within the hybrid workflow will outperform shallow algorithms. For testing our hypotheses, we will focus on the improvement of a forest RTM and the extraction of forest properties using advanced retrieval techniques, including spatio-temporal regularization, and hybrid methods with shallow and deep learning. For wall-to-wall mapping and validation activities, INFORM+ will combine EO satellite data (Sentinel-2, EnMAP and/or PRISMA, and GEDI) and airborne platforms with field campaign data. The innovative character of INFORM+ relates to the development of an improved forest RTM in the optical domain for generating LiDAR full-waveform signatures, along with multi- and hyperspectral signatures. We will implement spatio-temporal regularizations and identify suitable (deep) learning-based retrieval techniques to estimate the forest traits along with their uncertainty. Overall, INFORM+ will provide an update about forest properties in mid-European landscapes, allowing us to assess the potential of EO data with a suitable RTM to improve our understanding of forest ecosystems, thus bypassing the need for extensive ground data truth collection.

• DFG Programme: Research Grants
• International Connection: Austria, Finland, Luxembourg, United Kingdom
• Partner Organisation: Fonds National de la Recherche
• Cooperation Partners: Professor Dr. David Coomes; Dr. Markus Immitzer; Professorin Miina Rautiainen; Dr. Martin Schlerf
• Co-Investigator: Dr. Benjamin Brede