Skip to main content

3D-Monitoring of Ice and Forest from Space

Final Report Summary - 3D-MIFS (3D-Monitoring of Ice and Forest from Space)

1.1. Summary of the context and overall objectives of the project

In the current context of intense scientific discussion about the extent of climate change, the natural processes that capture more attention in the scientific community are the ones involved in the carbon cycle since they appear to be responsible for the most dramatic changes. Specifically, the Global Climate Observing System (GCOS) has identified a list of so-called Essential Climate Variables (ECVs) and, among them, those with a high impact (identified as “urgently needed data”): forest biomass change, deforestation, forest degradation and fires. The need to thoroughly monitor these processes is constantly rising. Better observation capabilities are required to reach a deeper understanding of Earth system dynamics and the impact the human activity is having on them. Nowadays, most of these processes are only selectively observed at low temporal and spatial resolutions. Regular and global scale observations are missing for many fundamental Earth system parameters such as vegetation biomass. Among the different tools for monitoring purposes, field measurements are sparse and non-uniformly distributed in space and time and they are restricted in providing data frequently enough at large scales. On the contrary, satellite observations can overcome these limitations and are potentially able to provide consistent temporal series in the long term, at global scale, with spatial resolutions and sampling frequencies in accordance with the spatial and temporal variability of the processes of interest. Therefore, ongoing research on Earth observation from space points to the successful exploitation of sensors capabilities for providing high quality geophysical products. In addition to the need of increasing spatial and temporal sampling, in the last years, attention is shifting more and more from 2D to 3D information by adding vertical structure observables, since it has been demonstrated that the horizontal distribution is not sufficient for an appropriate assessment of many physical parameters of interest.
In front of these needs, recent advances in spaceborne Synthetic Aperture Radars (SAR) offer unique and unprecedented opportunities both in terms of new acquisition configurations and 3D imaging. It has been shown that SAR signals at low frequencies can penetrate to a certain extent forest bodies and this capability allows, through advanced imaging techniques such as SAR tomography (TomoSAR in the following), the extraction of relevant information related to the 3D structural parameters of the observed scenes.

The objective of this project is two-fold and it aims to bridge the gap between the 3D observation capabilities of new SAR techniques and the development of novel bio/geophysical information products in the context of forest monitoring:

• The first objective is to design and evaluate the performance of a new imaging strategy for the retrieval and the monitoring of the dynamics of the 3D structure of forest bodies from multipass polarimetric and interferometric SAR data.

• The second objective relates to the physical interpretation of the obtained vertical reflectivity profiles for a range of forest types and conditions. This will lead to the derivation and evaluation of geophysical model functions to obtain an accurate retrieval of quantitative physical products of interest (vegetation structure, biomass, net primary productivity...) from these 3D structure measurements.

1.2. Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

The first half of the project was essentially devoted to fulfill the first objective dealing with designing a new imaging technique to obtain 3D reflectivity profiles from multibaseline polarimetric SAR acquisitions. A new approach based on Compressive Sensing theory was proposed and thoroughly tested on simulated and real data. The algorithm performs a full-rank tomographic inversion as far as fully polarimetric data is available (single rank in case of single polarimetric measurements). The technique has been compared to other state of the art techniques.

In parallel, attention was driven to understand how forest structure is reflected in the 3D reflectivity profiles obtained after the tomographic inversion processing step. First, through cross-check of airborne tomographic SAR measurements with field and Lidar data over a test site in Germany, the relevant forest structure elements to which tomographic SAR techniques at L-band are sensitive to were identified and a thorough study of the techniques regularly employed in forestry for forest structure estimation from field data was carried out. Since the measures commonly employed are based on characteristics of individual trees and remote sensing techniques cannot discriminate single trees, a direct translation of these terrestrial forest structure measures is not possible and indirect correspondences have to be explored instead. Taking this into account, two complementary forest structure descriptors were defined, one for the horizontal and one for the vertical structure. Their consistency with conventional measures of forest structure from field data was extensively verified on simulated scenarios.
Regarding real scenarios, since the 3D reflectivity profiles extracted from TomoSAR data are dependent of system configuration and scene characteristics, it is fundamental to check to which extent the descriptors proposed are influenced by these aspects. In order to do so, the method was applied to a temporal series of airborne TomoSAR data over a test site in Germany in parallel to a large set of simulations. The effect of non structural system and scene variabilities on forest structure estimation was analyzed, as well as its sensitivity to actual structural forest variations.
A major unexpected issue raised during the first half of the project was the difficulty encountered regarding validation of the method proposed in real scenarios. If only scattered local inventory plots are considered, the validation through field data is not complete, mainly due to the incompatibility of scales of the two observation systems. Besides the need of improving validation, in views of its application to data from future spaceborne systems at global scale, it is necessary to test on real scenarios the reproducibility of the approach in order to verify that very diverse stands in terms of structure are still comparable and that the method is sensitive to structure of forests at different latitudes. Hence, data from two major different campaigns were studied. The first one was proposed after discussions identifying the impossibility of establishing a link between remote sensing and field data when the scales of both sources are not compatible, together with the Chair for Forest Growth and Yield Science of the Technical University in Munich and with the Department of Ecological Modelling of the Helmholtz Center for Environmental Research in Leipzig. In order to address this issue, a test site in Germany was selected with a significant gradient in structure and a 25ha plot was established. More than 16000 trees were inventoried in early spring 2016 and TomoSAR data were acquired in two airborne campaigns in summer and in autumn 2016. The second campaign, called AfriSAR, took place in several selected locations in Gabon in February 2016 and gave the chance to check the behavior of the technique proposed for forest structure estimation in tropical forest stands from TomoSAR data at different frequency bands and from Lidar data.

1.3. Progress beyond the state of the art, expected results until the end of the project and potential impacts

The first result achieved in the framework of this project has been the development of a new method for TomoSAR imaging in distributed scattering scenarios. Its performance has been extensively compared to other state of the art techniques mainly in views of its application to forest structure retrieval. The assessment based on simulated datasets indicates that the method proposed achieves better sensitivity to forest structure and is more robust in front of temporal decorrelation. This could allow alleviating the requirements in terms of number and distribution of baselines of the TomoSAR system and this is extremely relevant in the system performance analyses of future space missions, such as Tandem-L.
Another main contribution of this project has been the proposal for the first time of a framework for the systematic quantitative forest structure estimation from TomoSAR data, based on ecology studies. It is intended to bridge the gap between individual based measures from field data and remote sensing observations in which the contributions of the different vegetation elements are aggregated in voxels and cannot be disentangled. The fact of reducing the dimensionality of the data when establishing two complementary forest structure descriptors from 3D observables allows a better comparison and characterization of the performance of different systems and for instance TomoSAR at L and P bands and Lidar data. This is essential in views of combining data from different sources in order to overcome individual limitations and increase the overall amount of information extracted, through data fusion techniques. And this is timely because in the near future, three main space missions are going to be launched mainly focused on biosphere applications, but based on different sensors: in 2018, the GEDI Lidar instrument will start acquiring measurements onboard the ISS; in 2021, the Biomass mission will do so with a SAR sensor at P band, as well as the Tandem-L mission with a SAR system at L band in 2020 and the NASA ISRO at L and S bands.
Also, a proper characterization of forest structure is a key proxy for a number of products of interest: accurate biomass estimation, forest productivity, identification of disturbances in the forest and assessment of their effects, among others. Therefore, the estimation of forest structure descriptors at global scales will allow a better assessment of the different processes in forest bodies.