Synthetic Aperture Radar (SAR) Tomography (TomoSAR) is a microwave imaging technology to recover the 3D structure of the illuminated scene by flying a Radar sensor along multiple trajectories. At microwave regime, electromagnetic waves are capable of penetrating into natural media that are non-transparent at optical frequencies, for example vegetation, snow, ice, and sand. This feature makes TomoSAR sensitive to the vertical structure of those media, hence providing a substantial advantage over optical sensors. The downside is that microwave scattering from distributed media may be quite complex, involving a number of different mechanisms through which the wave is backscattered. On forested areas, the Radar signal is determined by direct scattering from elements within the vegetation canopy, from the underlying terrain, as well as from multiple scattering resulting from the waves bouncing off the ground in the direction of the Radar after being scattered downwards by the tree canopy and trunks. The case of snow and ice bodies is even more complex, as signal backscatter is expected to arise from air/ice or snow/ice interface scattering, scattering from subsurface targets, bedrock, interfaces of layers with different dielectric permittivity, and eventually multiple scattering. In this context, the introduction of SAR Tomography techniques has opened up the way to a completely new approach to look at SAR data. A TomoSAR survey is typically obtained by repeatedly flying a Radar to acquire multiple SAR images of the same area from different trajectories. These data are then focused via digital signal processing algorithms to produce a collection of voxels that represent the backscattered energy in three dimensions, thus providing the possibility to see how the vertical structure of the imaged media interacts with the wave. Several papers have appeared in the literature reporting scientific analysis of natural media based on tomographic imaging, convincingly demonstrating of the uniqueness of this kind measurements. In the case of tropical forests, the use of tomography allowed the definition of new improved techniques for the retrieval of forest biomass and terrain topography under the vegetation layer. High-resolution tomographic surveys at X- and Ku-Band were used to study the vertical structure of snow and ice, revealing the presence interfaces between layers with different physical properties at the scale of few centimeters. Airborne L-Band tomography was used to study the interior of a glacier at the scale of meters, revealing details of its inner structure and allowing the detection of the bedrock at about 50 m below the ice surface.
The aim of this talk is to introduce tomographic SAR imaging to the audience. The talk will consider basic imaging principles and processing aspects, applications, and will discuss relevant case studies from campaign data over forest, snow, and ice.

The wavelength of synthetic aperture radar (SAR) waves affects the sensitivity to scatterers depending on their size and the penetration through forest volumes. Therefore, forest backscattering at different wavelengths (from P- up to X-band) contains different structural information about the geometry of vegetation elements and their distribution in space. On the other hand, the acquisition of more than one image in polarimetric interferometric (Pol-InSAR) or tomographic (TomoSAR) multi-baseline modes enables the estimation of structure parameters and 3-D reflectivity profiles, which in turn allow to access directly 3-D physical structure information.
An enhanced and accurate characterization of forest structure is expected to be achieved if Pol-InSAR and/or TomoSAR structure measurements at multiple frequencies are combined together based on their complementarity of information content. One way to understand structural complementarities and (possibly) develop combination algorithms is to refer to a “structure-based” framework. Although in a very early stage of development, it consists in establishing a correspondence between measurements at different frequencies by means of their capability in reflecting physical 3-D forest structure. This can be accomplished for instance by defining two complementary structure indices expressing heterogeneity in the horizontal and in the vertical direction with the aim to reflect the behavior of analogous indices based on ground inventory single-tree measurements.
In this presentation we first discuss and compare the information content of TomoSAR vertical profiles at multiple frequencies for different forest types. Then, the structure-based framework is reviewed and validated against experimental data (including Lidar and ground inventory measurements) acquired over different temperate and tropical forest conditions. The potentials and challenges of such framework to link structure measurements at multiple frequencies are addressed as well. Results achieved in the framework of actual multi-frequency multi-baseline airborne SAR campaigns (DLR F-SAR platform) in temperate and tropical forest campaigns in different configurations are reported and critically discussed.

The exploitation of satellite-borne remote sensing provided the opportunity to move from local to global scale in the study of forested environments. Among other sensors, SAR imaging looks promising as it offers continuous monitoring regardless of weather conditions or nighttime with fast revisit time and excellent resolution. Furthermore, the possibility of tuning the carrier frequency gives the opportunity to enhance the sensitivity to certain features neglecting others at the same time. Low frequency SARs, due to the significant penetration of the electromagnetic wave they use, they can image the whole forest layer, from the canopy top down to the ground level. This feature is particularly important for above ground biomass (AGB) estimation as the whole tree contributes to the radar echo making it an excellent proxy for AGB retrieval. The downside of such a complete imaging is that many different scattering mechanisms step in at the same time, mainly: direct return from canopies, from the ground and double bounces on tree parts and ground. The recorded signal in each range-azimuth resolution cell results from the coherent sum of many backscattered signals: each one associated with a physical process and dependent on the total AGB in different ways. The radar echo coming from the ground level is particularly illustrative: it carries information about the forest structure as the wave travelled through the whole forest layer before being recorded, however it also depends on the interaction with the ground itself. At P-band the ground echo changes with soil moisture and roughness but also with ground slope due to the fast disappearance of the double bounce scattering mechanism as the ground tilt drives away from 0 degrees. The problem of separating the scattering mechanisms involved was tackled by polarimetry and more recently by SAR tomography (TomoSAR). TomoSAR relies on the joint exploitation of a stack of coherent SAR images to focus on specific depths inside the vegetation layer: the capability of rejecting surrounding contributions is linked to the number of images at disposal and their spatial baselines. The exploitation of P-band TomoSAR provided a breakthrough in the estimation of tropical AGB: it resulted that the backscattered power at 30m above the ground level shows an extremely high correlation with AGB (more than 0.9). This result was obtained by Dinh et al. by analyzing the multi-polarimetric multi-baseline SAR data stack acquired by ONERA in 2009 on the forest surrounding Paracou (French Guiana) in occasion of the TropiSAR campaign. Main explanations for such a high correlation involve the biophysical structure of the tree or the features of the tomographic processing. From a biophysical point of view, in the surrounding of 30m above the ground level most of the biggest branches sprout, many leaves and small branches are also present. It covers the trunk extension only marginally so that the tree part that gathers most of the AGB is not properly imaged. Still, simulations showed that the percentage of the AGB in correspondence of 30m (about 30%) is constant with AGB itself making it an excellent proxy for AGB retrieval. The exceptionality of this elevation was discovered by the community working with LiDAR data too. They put into relation the area occupied by the tree at different elevation with AGB and found that the peak of the correlation was around 30m for different forests in different regions of the world. The second explanation for the high correlation between TomoSAR power at 30m and AGB has to do with the 3D focusing capability of the tomographic processing. SAR tomography linearly combines several SLC images with suitable weights emphasizing the signal coming from a specific elevation whereas surrounding contributions get attenuated. The attenuation depends on the geometry of acquisition and in the case of TropiSAR the rejection of the ground echo is very strong still maintaining the focus inside of the vegetation layer. According to this, the key feature of the tomographic processing might be the removal of the signal coming from the ground; removal of a signal that for sure carries information about the tree but it is very difficult to model. This possible reason motivated the processing chain here presented: the interferometric ground notching. As a matter of fact, the constraint of having at disposal a whole stack of SAR images to do tomography can be relaxed up to a single pair still achieving the ground cancellation. In order to do this the two images must be phase calibrated and phase flattened so that the interferometric phase associated with the terrain level is zero everywhere in the image; then the ground notched image is obtained as the difference between the two. The effect of this simple approach is to weight the vertical profile of the scattering target with a sinusoid function: the zero is placed at the terrain level thus achieving the ground cancellation, the peak is found in correspondence of half of the height of ambiguity (dependent, in turn, on the perpendicular baseline). A proper tuning of the height of ambiguity makes this interferometric approach analogous to tomography except for a smoother vertical impulse response function. The strong reduction of the number of images makes this approach suitable for interferometric or bistatic systems and greatly reduces the constraints on temporal decorrelation for repeat-pass tomographic systems (as for the forthcoming ESA BIOMASS mission); for these systems, it can also reduce the time interval between biomass estimates resulting in a more efficient way of processing the tomographic stack. In analogy with the signal returned by the tomographic processing, the strength associated with the ground notched image was put into relation with the local AGB: good correlation and sensitivity resulted.
This work presents the ground notching processing technique. Attention will be paid to the recent discoveries and evidences in AGB retrieval that led to the design of this algorithm; some technical details will be provided (as the required accuracy of the DTM or specific issues related to airborne surveys); finally, the correlation with AGB of the ground notched power will be shown.

Boreal forests store nearly one third of global terrestrial carbon and are considered one of the highest soil carbon dense region in the world. In 2017-2018 NASA conducted 36 tomographic flights as part of the Arctic-Boreal Vulnerability Experiment (ABoVE) over Delta Junction in Alaska with the Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) complemented by (1) full-waveform lidar acquisitions with the Land, Vegetation, and Ice Sensor (LVIS) instrument, (2) small-footprint lidar acquisitions with the Lidar, Hyperspectral and Thermal (G-LiHT) airborne imager and (3) field data collection. Additional UAVSAR flights were conducted in coordination with the German Aerospace Center’s L/S-band F-SAR tomographic radar instrument over the Berms site in Alaska. The UAVSAR images were acquired using L and P bands, providing the opportunity for a systematic multi-frequency tomographic radar (TomoSAR) analysis in preparation for the L/S-band NISAR and P-band BIOMASS satellite missions.

The goal of this talk is to discuss our comparative assessment of multi-frequency UAVSAR tomographic profiles with lidar waveforms over Alaskan boreal forests, and how radar tomograms can be used to extract vegetation structural parameters to support ecosystem applications such as carbon modeling, vegetation disturbance mapping and biodiversity monitoring. The ABoVE campaign tomographic radar data have been processed through the same processing flow used for the 2016 AfriSAR campaign data, which includes tomographic stack formation with no spectral shift and filtering, tomographic phase calibration and vertical profile generation using the Capon and Fourier methods. L and P band tomographic stacks are registered to each other and with lidar samples after multi-looking to conduct an accurate analysis across the different frequencies and instruments.

TomoSAR-derived vertical profiles are further processed to extract ecosystem parameters. Several parameters are being considered and evaluated by comparing them with either field data or lidar-derived parameters. Mean canopy height and ground topography are two key parameters, but other metrics that describe the vertical structure (e.g., relative location and magnitude of the main tomogram/waveform peak) as well as the horizontal heterogeneity of the canopy (e.g., spatial standard deviation of the vertical parameters) will be examined. In our preliminary, we found that lidar waveform algorithms developed by the lidar community could be adapted and applied to radar tomograms accounting for the different nature of lidar and radar measurements. These algorithms could potentially include also regression models to estimate above-ground biomass from radar tomograms, providing new opportunities for fusing lidar- and TomoSAR-derived ecosystem parameters.

A tower-based experiment, named BorealScat, was set up in late 2016 for acquiring fully-polarimetric, tomographic radar images of a boreal forest site at P- and L-band. In 2017, the system was upgraded to include C-band measurements. Tomographic images have been acquired with negligible temporal decorrelation at five-minute intervals, resulting in a densely-sampled, multi-year tomographic time series of a boreal forest site. The goal of the experiment is to gain new knowledge of how synthetic aperture radar (SAR) measurements (e.g. backscatter and coherence) over boreal forests are affected by environmental variables (wind, temperature, rain etc.), what electromagnetic scattering mechanisms take place during such measurements and how these measurements differ over the vertical extent of the forest canopy. The results are applicable to forest variable (e.g. biomass, height and moisture) estimation algorithm design using SAR data as well as future airborne and spaceborne SAR mission design.

The experiment consists of a 50 m-high tower equipped with a radar system including an array of 30 antennas. The system was designed and constructed for acquiring fully-polarimetric tomographic images at P-, L- and C-band with high temporal and spatial resolution. BorealScat’s L- and C-band measurements operate at higher radar frequencies than previous ground-based experiments, presenting new challenges in avoiding temporal decorrelation during measurements. This tower-based experiment is exposed to the environmental elements, has a limited field of view and acquires images in a complex 3D geometry, making the system design and calibration particularly challenging. Novel calibration and processing methods have been developed for producing high-quality products from BorealScat’s measurements.

The time series span several seasons, revealing how radar measurements over boreal forests are affected by freeze-thaw cycles during the winter and diurnally-varying transpiration rates during the summer. On the shortest timescale, wind has been observed to cause fluctuating radar backscatter, with increasing sensitivity for the higher frequencies, and even a backscatter bias at P-band. On the longest timescales, soil moisture changes affect the ground-level backscatter in P-band tomographic images. Strong rainfall caused transients in the radar observables, but light rain is unlikely to have any noticeable effect at P-, L- or C-band. The dataset has been studied to reveal the optimal SAR acquisition times and seasons for minimizing repeat-pass temporal decorrelation at P- and L-band. C-band average backscatter power was found to be much less sensitive to temporal variations than P- and L-band, showing promise for proposed future companion satellites to Sentinel-1.

The knowledge gained from BorealScat’s results have allowed insight into how electromagnetic scattering models and forest variable estimation can be improved. Some schemes have been suggested for mitigating the effects of environmental variables on SAR data.

One of the key recommendations resulting from previous radar-based earth observation mission concepts for snow parameter retrieval is that high-resolution imagery of the vertical structure of snow is required to gain further insight into the complex electromagnetic interaction within snowpacks.
As part of the ESA SnowLab campaign the SnowScat device, a terrestrial stepped-frequency continuous-wave (SFCW) scatterometer, which supports fully-polarimetric measurements within a frequency band from 9.2 to 17.8 GHz,
was operated in tomographic profiling mode to obtain time series of high-resolution vertical snow profiles.
In this tomographic profiling mode, the SnowScat device is subsequently displaced in elevation direction to obtain a high resolution not only in range direction but also along elevation. This leads to two-dimensional vertical profiles of the snowpack, which means that observables such as radar backscatter, co-polar phase difference, interferometric phase and coherence can be distinguished also along the vertical dimension.
In this talk, (1) the measurement and imaging concept of SnowScat in tomographic mode is presented, (2) examples of acquired data products obtained within the ESA SnowLab campaign in the Swiss Alps during four snow seasons are shown, and (3) approaches to retrieve snow parameters as well as comparisons with in-situ snow measurements are discussed.
In particular, these time series of tomographic profiles and accompanying in-situ snow characterizations have allowed to investigate different phenomena, such as using the variation of radar backscatter to locate melt-freeze crusts/horizontal
layers within the snow pack, or using the co-polar (HHVV) phase difference to characterize potential anisotropy or changes in anisotropy, or using differential (temporal) coherence between tomographic profiles to assess temporal changes in the propagation delay throughout the time series, spatially resolved in the 2-D vertical profile.
The last three snow seasons contained different weather regarding time of the first snow fall, temporal distribution of snow fall and temperature, and a total snow accumulation with a maximum of more than 2m snow depth at the test site Laret in Davos, Switzerland, in January 2018.
The ESA SnowLab is operated again in the upcoming winter season 2018/2019 to continue the successful time series measurements which means that latest results of this winter are anticipated to be included in the LPS2019 presentation.