Fringe 2017 > Session details
Paper 10 - Session title: Thematic mapping, vegetation and DEMs
11:10 A Global Forest/Non-Forest Map from TanDEM-X Interferometric Data
Rizzoli, Paola; Martone, Michele; Wecklich, Christopher; Gonzales, Carolina; Bueso Bello, Jose´ Luis; Krieger, Gerhard; Zink, Manfred German Aerospace Center, Germany
The global interferometric SAR data set provided by the TanDEM-X mission represents a highly valuable source for many scientific applications, among them land classification. In particular, the identification and monitoring of vegetated areas plays a key role in a large variety of different fields, such as agriculture, cartography, geology, forestry, global change research, and regional planning.
In this paper we present our activities for generating a global forest/non-forest map starting from TanDEM-X interferometric SAR data, acquired by the TerraSAR-X and TanDEM-X satellites in bistatic configuration. The algorithm applied to each individual TanDEM-X interferometric acquisition exploits the decorrelation contribution due to volume scattering. Volume decorrelation represents the interferometric coherence loss which is caused by multiple scattering within a volume and is a well pronounced indicator of the presence of vegetation. Given a coherence estimate, it is possible to isolate the volume decorrelation contribution and this characteristic can be used to distinguish vegetated from non-vegetated areas. A weighted clustering approach based on fuzzy logic is utilized for partitioning each pixel into two classes: forest or non-forest, by associating a membership value to it, which expresses the weighted probability of an observation to belong to each single class.
Furthermore, several overlapping coverages are globally available and have to be properly combined together to generate consistent large scale maps. The method for the mosaicking of overlapping scenes will also be presented. It is based on a weighted average of multiple membership values which takes into account indicators of their reliability, such as the dependency of volume decorrelation on the height of ambiguity, on the signal-to-noise ratio (SNR), and on the acquisition geometry. For example, the inclusion of the SNR component in the weighting logic reduces the influence of pixels located on the edge of each scene where the SNR is reduced due to a lower antenna gain. Various additional layers are derived and applied on top of the forest/non-forest map, for example urban areas and water bodies are removed, before the input membership values are finally mosaicked together and quantized into a binary value. The delivered product will be characterized by a resolution of 50 m x 50 m and global coverage.
The validation and performance measurement approach using ground truth maps and independent sources will be described and preliminary results will be discussed as well.
Finally, if input data covering a certain time span are available, the developed method can be used to detect temporal changes in the vegetation coverage. To conclude, we will present some examples for detecting on-going deforestation in the Amazon rain forest.
Paper 47 - Session title: Thematic mapping, vegetation and DEMs
12:10 InSAR Forest/Non-Forest Classification Exploiting Nonlocal Pixel Similarities
Sica, Francescopaolo; Martone, Michele; Rizzoli, Paola DLR - Deutsches Zentrum für Luft- und Raum-fahrt, Germany
Mapping of forested areas around the globe and monitoring their changes throughout the years are challenging tasks that can be effectively accomplished by means of Earth observation satellites. In particular, interferometric SAR acquisitions can be exploited for the generation of both local and global forest/non-forest maps. The principle is to exploit the decorrelation contribution due to volume scattering, which results from the penetration of the radar wave within the canopy. This effect depends on several parameters, such as the forest vertical profile and its density, the sensor frequency, and the viewing geometry. It is named volume decorrelation and can be extracted from the total interferometric coherence. Given its clear correlation to the presence of vegetation, it represents a highly valuable input in simple classifiers for discriminating forested areas.
In this work, we investigate the use of nonlocal filtering for preserving data resolution and features and improve the quality of the estimated coherence. Up to now, nonlocal filtering has been successfully used in SAR intensity and interferometric parameters estimation. The use of this approach for classification purposes represents a promising way to fully exploit the capabilities of interferometric SAR systems. The nonlocal principle consists of looking for similar pixels in a given search area and perform a weighted average according to their grade of similarity. Depending on the weight distribution, the equivalent number of looks (ENL) varies within a single scene, allowing in this way to preserve small features and hence the spatial resolution. Moreover, the coherence estimation is biased especially for lower coherence values, which in most cases refer to forested areas because of volume decorrelation. Hence, in order to have an almost constant bias error on coherence-homogeneous areas, the implemented nonlocal method forces the filtering to a fixed ENL with a consequent marginal detriment of the resolution. This nonlocal method is usually named adaptive multilooking.
We apply nonlocal filtering to both Sentinel-1a/b constellation and bistatic TanDEM-X data and, starting from the estimated volume decorrelation, we then discriminate forested areas by applying a clustering method based on fuzzy logic. Preliminary results on the Amazon forest show the improved resolution preservation and, even more important for classification purposes, a better separation in terms of decorrelation values between forested areas and non-forested ones. Further developments concern the use of the nonlocal pixel similarity in the classification procedure itself and will be investigated in the near future.
Paper 286 - Session title: Thematic mapping, vegetation and DEMs
11:50 Large-scale mapping of forest standing volume with interferometric X-band SAR
Solberg, Svein; May, Johannes Norwegian Institute of Bioeconomy Research, Norway
Background: Tandem-X has provided wall-to-wall coverage of single-pass InSAR data pairs over Norway a number of times, and this opens up for developing a wall-to-wall forest resource map over Norway. Through a number of earlier studies in smaller areas we have demonstrated the strong and fairly stable relationship between the interferometric phase height above ground and standing volume. Important infrastructures which are largely in place enable this: A considerable fraction of the productive forest area has been covered by airborne laser scanning, and all the National Forest Inventory (NFI) plots are accurately positioned to be used for model calibration and validation. However, there are a number of issues to be solved prior to this, including clarifying the effect of between-dataset variations in dielectric properties due to weather, what to do in areas without an accurate DTM and model selection.
Aim: The aim with this paper was to develop a method to map standing volume throughout Norway, based on detailed research in one selected county.
Materials and methods: We selected the county of Østfold in the Southeast corner of Norway, with an area of 4000 km2. This county had recently a special county forest inventory, providing additional field data and accurate, aggregated statistics. In November 2015 this county was covered by 9 TanDEM X acquisitions. We processed the CoSSC data into a Digital Surface Model (DSM) with 10 x 10 m spatial resolution for each acquisition, and mosaicked them together to a seamless DSM with full coverage. In addition we obtained the local coherence for the same resolution by using a local 5 x 5 window around each cell. The entire county was covered by airborne laser scanning in 2015, and we obtained an accurate DTM with 10 x 10 m spatial resolution. We subtracted the DTM from the DSM and obtained InSAR height with full coverage, representing a canopy height model. We established a ground truth data set from the NFI plots, together with an equal amount of ad-hoc field plots having the same layout. The total number of plots was 600, of which 300 were ordinary NFI plots. All field plots were circular and 250 m2, and the field measurements included tree species, diameter at breast height (DBH), and height on a number of trees. The volume of each tree was obtained from Norwegian allometric models. For each field plot we selected one 10 x 10 m InSAR height- and coherence data sets, i.e. the cell being nearest to the field plot center. Individual forest volume models were derived from the InSAR data using a linear regression. For InSAR height we used no-intercept analysis-of-covariance. This was done for different tree species, all acquisitions separately and all acquisitions mosaicked. A tree species map with the classes pine, spruce and broadleaves ere generated with Sentinel-2 data trained with the NFI plots.
Results: Preliminary results showed a small difference between spruce and pine, however, not statistically different at the α=0.05 level. The analyses are currently in progress, and more results will be produced prior to the conference. We will compare simple linear models with biophysical models like Random Volume over ground (RVoG) and the two-level model (TLM).
Paper 382 - Session title: Thematic mapping, vegetation and DEMs
12:30 TanDEM-X for national forest mapping
Persson, Henrik J (1); Nilsson, Mats (1); Olsson, Håkan (1); Fransson, Johan E.S: (1); Soja, Maciej J (2); Ulander, Lars (2) 1: Swedish University of Agricultural Sciences, Sweden; 2: Chalmers University of Technology, Sweden
Many developed countries including Sweden have national coverages with airborne laser scanning (ALS), performed at least once, from which an accurate digital terrain model (DTM) can be constructed. Moreover, the forest sector can benefit from wall-to-wall remote sensing data that has proven to give accurate estimations of common forest attributes, such as tree height, basal area and stem volume. However, this information gets outdated quickly, because of constant changes in the forest, caused by for example storm hazards, thinning and cuttings, growth and so on. One alternative to update the Swedish forest map product Skogliga Grunddata(SGD), which is based on ALS data, is to use satellite based techniques which can provide frequent acquisitions at sufficient resolution.
In this study, we have used a nation-wide collection of about 500 TanDEM-X images, the ALS based DTM, and thousands of 7 to 10 m sample plots from the Swedish National Forest Inventory (NFI) in order to create a nation-wide pixel product. The possible update or replacement of existing ALS based products is partly evaluated and the preliminary results are pointing to this method having great potential, however having several important challenges to be overcome, before an automatically generated product can be delivered. This includes suitable acquisitions, correct phase unwrapping, relevant height calibration, and moreover a meaningful merging of the TanDEM-X tiles.
In summary, the potential of using X-band radar data for frequent boreal forest mapping appears high, accounted that some crucial challenges are overcome.
Paper 487 - Session title: Thematic mapping, vegetation and DEMs
11:30 Forest Parameter Estimation Using New Semi-empirical InSAR Coherence Models
Praks, Jaan (1); Antropov, Oleg (1,2); Olesk, Aire (3); Voormansik, Kaupo (3) 1: Department of Electronics and Nanoengineering, Aalto University, P.O. Box 15500, 00076 AALTO, Finland; 2: VTT Technical Research Centre of Finland, P.O. Box 1000, 02044 VTT, Finland; 3: Department of Space Technology, Tartu Observatory, 61602 T ̃oravere, Tartumaa, Estonia
Rapidly developing Synthetic Aperture Radar interferometric techniques have lately provided new possibilities for accurate forest remote sensing with existing short wavelength spaceborne SAR systems. In this work we assess the accuracy of forest height estimation technique based on inversion of recently published semi-empirical coherence models  and TanDEM-X single pass interferometric images.
Currently the X- and C-band SAR systems are the most common SAR sensors in space, because of their optimal antenna size and resolution combination for spaceborne use. Unfortunately these wavelengths are not considered ideal for remote sensing of world forests, although the relation between forest biomass and backscattered signal is well known. Utilization of theses systems is hindered by the fact that at X- and C-band the backscattering-forest biomass relation saturates rapidly for X- and C-band systems, making it difficult to use for mature forest. The result is also strongly dependent on imaging conditions.
Luckily, recent advancements and new techniques, such as SAR polarimetry and interferometry, have opened new opportunities for forest remote sensing also with shorter wavelengths. This opens entirely new utilization area for example for TanDEM-X mission measurements. It has been shown, that polarimetric single pass interferometric measurements combined with advanced coherence models allow model inversion which can produce also rather accurate forest height estimate. It is also shown, that this technique works for surprisingly short wavelengths and allows height estimation even for X-band high resolution SAR . Unfortunately, the PolInSAR technique requires fully polarimetric data and/or accurate knowledge of local topography, which is not widely available. Moreover, the model inversion involves complicated techniques, not applicable for large areas in practical usage.
The latest results by our team  show, that the complexity of PolInSAR coherence models can be significantly reduced without significant sacrifice of the accuracy when some ancillary information about forest type and imaging conditions are available. It was shown in  that X-band InSAR coherence in winter conditions can be accurately predicted based on lidar measured forest height. This can be done even with simple single parameter models and single polarization SAR measurement . It was also demonstrated that proposed models can be easily inverted for forest height, forming a toolbox for large area tree height estimation from TanDEM-X coherence images in boreal and hemi-boreal region.
In this work we analyse in more detail the inversion problem of proposed semi-empirical models, where the tree height is derived directly from X-band InSAR single pass coherence images. The multi-temporal set of several TanDEM-X interferometric pairs and the 90th percentile lidar based forest height maps over test area in Estonia are used in this study. Various scenarios in terms of available ancillary data and model complexity are analysed. It is shown, that additional information of tree species can improve tree height estimation accuracy and the winter scenes provide the most accurate tree height estimation. It is shown, that for homogeneous pine stands, the height estimation accuracy approaches even lidar measurement accuracy, however, additional training data for the model is required. Whether the information on forest species is missing, the model inversion can still be applied with slight sacrifice of the overall accuracy, but the method still provides far better accuracies than backscattering amplitude based methods in high spatial resolution. We also propose some ideas how this method can be adopted to Sentinel-1 interferometric images and give some leads how the temporal decorrelation problem could be solved.
 A. Olesk, J. Praks, O. Antropov, K. Zalite, T. Arum ̈ae ja K. Voormansik. Interferometric SAR Coherence Models for Characterization of Hemiboreal Forests Using TanDEM-X Data. Remote Sensing 8.9 (2016), s. 700. issn: 2072-4292. doi: 10. 3390/rs8090700. url: http://www.mdpi.com/2072-4292/8/9/700.
 J. Praks, A. Olesk, K. Voormansik, O. Antropov, K. Zalite ja M. Noorma. Building Blocks for Semi-empirical Models for Forest Parameter Extraction from Interferometric X-band SAR Images. Teoksessa: IGARSS 2016, July 10-15, 2016, Beijing, China. 2016.
 J. Praks, O. Antropov ja M. T. Hallikainen. LIDAR-Aided SAR Interferometry Studies in Boreal Forest: Scat- tering Phase Center and Extinction Coecient at X- and L-Band. IEEE Transactions on Geoscience and Remote´Sensing 50.10 (2012), s. 38313843. issn: 0196-2892. doi: 10.1109/TGRS.2012.2185803.