Speaker
Description
Wenyu Yang [1] , H. Aghababaei [2], Giampaolo Ferraioli [1]
[1] Dipartimento di Ingegneria, Università degli Studi di Napoli “Parthenope,” 80143, Naples, Italy
[2] University of Twente, Faculty of Geo-Information Science and Earth Observation
Enschede, The Netherlands
Characterizing and monitoring forests are essential for tracking climate change and quantifying the global carbon cycle, particularly through aboveground biomass (AGB) mapping. This involves identifying their vertical structure, representing tree, trunk, and crown arrangement—an important productivity and biomass indicator recognized by the scientific community [1]. Traditionally, forest monitoring requires costly and time-intensive in situ fieldwork, yielding precise data. In contrast, remote sensing methods offer widespread, frequent coverage. Techniques ensuring under-foliage penetration and 3D measurements, such as high-resolution waveform lidar and Synthetic Aperture Radar (SAR) tomography (TomoSAR) [2], are ideal for forest monitoring.
SAR stands out due to its weather independence and broad coverage. Yet, SAR images project 3D data into a 2D domain. TomoSAR solves this issue by synthesizing a vertical array, enabling full 3D imaging. TomoSAR is indeed a unique tool for accurate 3D forest reconstruction, delivering reliable biophysical data like vertical structure and biomass, akin to terrestrial and lidar approaches.
Various standard spectral estimation techniques, such as Beamforming and Capon, have been employed to reconstruct the distribution of vertical backscattering from multi-baseline SAR images of forested regions [3]. Results from numerous airborne campaigns, including BioSAR [4], AFRISAR [5], and TropiSAR [6], have demonstrated the effectiveness of these methods. This progress indicates a promising direction for deriving and reconstructing 3D backscattering images of forested areas from spaceborne data.
Currently, several missions, including BIOMASS [7], NISAR[8], and the Tandem-L SAR [9], are scheduled for launch in the next two years. Utilizing low-frequency images from these systems provides optimal SAR images for in-depth examination of forested areas due to their high signal penetration capabilities. However, a key distinction between upcoming spaceborne data and existing airborne multi-baseline SAR images lies in the baseline distribution of the sensors, temporal decorrelation and atmospheric effects that can affect quality of tomographic inversion. Recently, deep learning based methods have become a fundamental methodology for different remote sensing image processing tasks as well as for SAR image processing [10-11]. This study aims to take the advantages of deep learning to improve forest height estimation in tomographic framework. Observing the estimation of the forest height from a classification point of view, a fully connected network is designed to extract the abstract features for height retrieval over covariance matrix of tomographic data along with the LiDAR data as the reference. The main and noteworthy outcome of the proposed solution lies in its ability to generate a resilient ground and canopy height model, irrespective of the presence or absence of calibration phase errors (atmospheric errors) in the images.
References
[1] Hardiman BS, Bohrer G, Gough CM, Vogel CS, Curtisi PS. The role of canopy structural complexity in wood net primary production of a maturing northern deciduous forest. Ecology. 2011 Sep;92(9):1818-27. doi: 10.1890/10-2192.1. PMID: 21939078.
[2] Aghababaei, H., Ferraioli, G., Ferro-Famil, L., Huang, Y., d'Alessandro, M. M., Pascazio, V., ... & Tebaldini, S. (2020). Forest SAR tomography: Principles and applications. IEEE geoscience and remote sensing magazine, 8(2), 30-45..
[3] Ferro-Famil, L., Huang, Y., & Ge, N. (2022, July). Forest structure characterization using SAR tomography and an adaptive estimation technique. In EUSAR 2022; 14th European Conference on Synthetic Aperture Radar (pp. 1-4). VDE.
[4] L. M. H. Ulander et al., "BIOSAR 2010 - A SAR campaign in support to the BIOMASS mission," 2011 IEEE International Geoscience and Remote Sensing Symposium, Vancouver, BC, Canada, 2011, pp. 1528-1531, doi: 10.1109/IGARSS.2011.6049359.
[5] Hajnsek, I., Pardini, M., Jäger, M., Horn, R., Kim, J. S., Jörg, H., ... & Wasik, V. (2017). Technical assistance for the development of airborne SAR and geophysical measurements during the AfriSAR experiment. ESA contract no. 4000114293/15/NL/CT.
[6] Dubois-Fernandez, P. C., Le Toan, T., Daniel, S., Oriot, H., Chave, J., Blanc, L., ... & Petit, M. (2012). The TropiSAR airborne campaign in French Guiana: Objectives, description, and observed temporal behavior of the backscatter signal. IEEE Transactions on Geoscience and Remote Sensing, 50(8), 3228-3241.
[7] Mission Overview - ESA Earth Online - European Space Agency, Biomass infographic, available at https://earth.esa.int/eogateway/missions/biomass/description
[8] P. Hoffman et al., "NASA-ISRO Synthetic Aperture Radar (NISAR) Mission: System Integration & Test," 2022 IEEE Aerospace Conference (AERO), Big Sky, MT, USA, 2022, pp. 1-17, doi: 10.1109/AERO53065.2022.9843829.
[9] Krieger, Gerhard, Irena Hajnsek, Konstantinos Papathanassiou, Michael Eineder, Marwan Younis, Francesco De Zan, Pau Prats et al. "The tandem-L mission proposal: Monitoring earth's dynamics with high resolution SAR interferometry." In 2009 IEEE Radar Conference, pp. 1-6. IEEE, 2009.
[10] X. X. Zhu et al., “Deep learning in remote sensing: A comprehensive review and list of resources,” IEEE Geosci. Remote Sens. Mag., vol. 5, no. 4, pp. 8–36, Dec. 2017.
[11] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” in Proc. Adv. Neural Inf. Process. Syst., 2012, pp. 1106–1114.
Summary
This topic pertains to SAR tomography and the efficient reconstruction of vertical images of forested areas using upcoming spaceborne SAR data.
