Speaker
Description
The TomoSense experiment was conceived to provide the scientific community with unprecedented data to study the features of radar scattering from temperate forests, comprising tomographic and fully polarimetric SAR surveys at P-, L-, and C-band, acquired in mono- and bistatic mode by simultaneously flying two aircraft. The TomoSense dataset is complemented by a detailed forest census, Terrestrial Laser Scanning (TLS), and Airborne Lidar Scanning (ALS) products.
All campaign activities were finally and successfully closed in fall 2021, resulting in an amount of over 1800 SAR images at different polarizations, frequency bands, and acquisition modes.
A significant part of data processing activities was dedicated to interferometric and tomographic calibration of SAR data, necessary to finely estimate platform motion and achieve accurate tomographic focusing Calibration activities resulted in the generation of finely coregisted, phase calibrated, and ground steered complex SAR image stacks at all frequency bands, which were included in the final data delivery to ESA. Calibrated image stacks were afterwards processed to generate multi-frequency mono- and bi-static Tomographic cubes representing forest scattering in three dimensions, intended to serve as the basis for all subsequent scientific analyses and also included in the final data delivery.
Scientific analyses were conducted by investigating the connection of Tomographic cubes to biophysical parameters. Essential to these activities was the availability of a large amount of biophysical information from independent measurements, including field-works, TLS, and ALS.
Based on the large amount of results produced by the study, the following conclusions are drawn.
P- and L-Band: Both P and L band are observed to provide sensitivity to the whole vegetation layer, in that both frequencies allow for a clear detection of terrain and forest canopies. Moreover, both frequencies are robust w.r.t. temporal decorrelation over few hours, resulting in the possibility to produce high-quality tomographic imaging from repeat-pass campaign data. Interestingly, bistatic data at L-Band are observed to contain weaker contributions from the terrain level than mono-static data. As a result, the Ground-to-Volume backscatter power ratio (G2V) is systematically larger for monostatic measurements by up to 4 dB, depending on polarization and local conditions (understory, topography). Forest height could be successfully estimated using repeat pass mono-static P- and L-Band data, repeat pass bi-static L-Band data, as well as using only simultaneous interferograms from each bistatic pass.
Interestingly, the use of normalized tomographic indicators like fractional volume intensity was observed to provide sensitivity to AGB as well, the best result being assessed in slightly over 20% on the aggregated forest class using bistatic L-Band data.
C-Band: Tomographic analysis of C-Band data indicates that the residual coherence in repeat-pass interferograms is mostly determined by scattering from the ground level, whereas the signal from the forest canopy is nearly impossible to detect because of temporal decorrelation. Analysis of interferograms formed by mono- and bi-static data collected in the same flight reveals that tall forests decorrelate almost immediately because of wind gusts, thus confirming the results from the BorealScat experiment. However, tomographic processing allowed for exploiting vestigial coherence from the vegetation, resulting in the possibility to detect forest canopies in parts of the image and observe a good match w.r.t. Lidar height.
Overall, results obtained at this site indicate that C-Band waves care capable of penetrating down to the ground level. This finding provides an element in support of the feasibility of C-Band tomography of temperate forests, clearly provided that acquisitions are taken at a temporal baseline of a few tens of milliseconds.
The TomoSense data-base: one important product of the study consists in the creation of a data-base intended to serve as an important basis for studies on microwave scattering from forested areas in the context of future studies on Earth Observation missions. The data-base comprises complex SAR images and tomographic cubes at different levels of processing, as well as ALS-derived maps of forest height and AGB, forest census, and TLS profiles.
Complex SAR images in the data-base are already finely coregistered, phase calibrated, and ground steered, in such a way as to enable future researchers to directly implemented any kind of interferometric or tomographic processing without having to deal with the subtleties of airborne SAR data. In addition to that, the data-base comprises tomographic cubes representing forest scattering in 3D both in Radar and geographical coordinates, which are intended for use by non-Radar experts.
Overall, we emphasize that the amplitude of the TomoSense data-base is such that a number of questions had to be left unexplored within the time of the study, and in particular those concerning the joint use of different tomographic observables. For this reason, it is our opinion that the TomoSense data-base will represent a most valuable tool for student and researchers in the next years.
