• 12:30 Above Ground Biomass estimation from SAR vertical reflectivity profiles

  Toraño Caicoya, Astor (1); Kugler, Florian (1); Pardini, Matteo (1); Hajnsek, Irena (2); Papathanassiou, Konstantinos (1) 1: German Aerospace Center (DLR), Germany; 2: ETH Zürich

  Show abstract

  First attempts in the estimation of above ground forest biomass (AGB) by means of SAR, have used allometric equations to calculate biomass from the estimated PolInSAR forest height. However, changes in forest density or structure bias the known allometric relations. Today, remote sensing systems like SAR are able to measure additional parameters like vertical forest structure [1] [2]. In this work, the potential of vertical forest structure information, estimated from SAR reflectivity profiles, will be investigated for the improvement of biomass inversions. The structure characterization developed here is based on the principles proposed in [3]. These are explained as follows: high biomass forest stands are characterized by frequencies connected to tree stems approaching a homogeneous biomass distribution. However, for the same dominant height (H100) a stand with lower biomass, presents more gaps, and characterized by a higher proportion of frequencies connected to the crown compartments. Using the Legendre decomposition we propose a structure ratio (Srat) to characterize the vertical biomass distribution, which is based on the proportion between stem (a01), and crown frequencies (a02+a03+a04): Srat = a01/( a02+a03+a04). In this work, airborne L-band data acquired with DLR’s E-SAR during the TempoSAR 2008 campaign are used to investigate the capabilities of SAR tomography to invert forest biomass. Data were acquired on two days, one before a rain event and another after. Each dataset used is composed of 5 images (baselines 5, 10, 15, 25m). The Capon method is applied on phase calibrated images to derive the vertical reflectivity profiles [4]. The resulting image has a horizontal spatial resolution of ~12x12m. The biomass inversion algorithm proposed here comprises three stages: a profile based height truncation, an extinction correction and the application of an empirically derived structure-to-biomass allometric relation. Profile height detection, at which the profile will be truncated, is based on the distribution of the profile lobes, i.e. local profile maxima. It is possible that a vertical profile presents several physically relevant lobes [5], so the estimation of the height from vertical reflectivity profiles requires distinguishing between physically relevant lobes and side-lobes. As the ground location is known [4] the detection of the side-lobes is only necessary above the physically relevant lobes. Then, the reflectivity profile is truncated when the power amplitude of the last lobe with a power larger than a 20% of the absolute maximum power, is below a 30%. Reflectivity profiles do not represent the exact vertical biomass distribution as obtained from ground measurements. Therefore, in order to apply the structure ratio, the reflectivity profiles need to be corrected. We have observed that SAR reflectivity profiles the range of the ratio is strongly reduced, due to the typical overemphasis of the power of the highest profile maxima induced by the signal extinction along forest height. In order to compensate for this effect, an exponential correction is applied for every profile. This correction is adapted for each profile, depending on the profile height and the proportion between the main profile maxima. The biomass inversion needs to be adapted to the information contained in the SAR vertical reflectivity profiles. As introduced before, the low frequency coefficient a01 is mainly driven by the biomass concentration in the tree stems and long wavelengths, like L-band, are not so sensitive to them. As a consequence, a higher presence of biomass will be indicated by larger crown volumes along the profile and therefore by a higher proportion of crown frequencies (a02, a03, a04) in the structure ratio. This will be translated in a change of the Srat exponent sign in the biomass inversion. Inversion details will be shown. Finally, we will show that the proposed structure-to-biomass allometry is able to improve the biomass inversion results obtained with height allometries in more than a 20%, in terms of correlation coefficient and RMSE, showing the potential of low frequency SAR data (L-band) for the estimation of AGB. The structure descriptor based on Legendre polynomials is able to characterize the vertical reflectivity profiles and successfully connect this information to forest biomass. The algorithm is tested for wet and dry forest conditions, as these can affect the shape of the reflectivity profiles, and a sensitivity analysis will evaluate the potential of this technique for biomass classification. [1] K. Papathanassiou and S.R. Cloude, “Single-baseline Polarimetric SAR Interferometry”, IEEE Transactions on Geoscience and Remote Sensing, pp. 2352-2363, 2001. [2] R.N. Treuhaft et al., “Vegetation profiles in tropical forests from multibaseline interferometric synthetic aperture radar, field, and lidar measurements”, Journal of Geophysical Research. pp. 114, 2009 [3] A. C. Torano, F. Kugler, I. Hajnsek, K. Papathanassiou,, “Vertical forest structure characterization for the estimation of Above Ground Biomass. Potential and limitations for Radar Remote Sensing”, Proc.of EUSAR, Berlin, 2014. [4] M. Pardini, A. Cantini, F. Lombardini and K. Papathanassiou, “3-D Structure of Forests: First Analysis of Tomogram Changes Due to Weather and Seasonal Effects at L-Band”, Proc. of EUSAR, Berlin, Jun 2014 [5] S. Tebaldini and F. Rocca, “Multibaseline Polarimetric SAR Tomography of a Boreal Forest at P- and L-Bands”, IEEE Transactions on Geoscience and Remote Sensing, vol 50., no 1. pp. 2352-2363, 2012

  
   

• 12:50 Vegetation height estimation performance with respect to RVoG model deviations for full and compact polarimetry

  Arnaubec, Aurélien (1,2); Dubois-Fernandez, Pascale (1); Roueff, Antoine (2); Koleck, Thierry (3) 1: ONERA; 2: Ecole Centrale Marseille, France; 3: CNES

  Show abstract

  Introduction The Random Volume over Ground (RVoG) is a widely used model in PolInSAR analysis [1]. It has been shown that it can be interesting to analyze the precision of vegetation and ground heights with the Cramer–Rao bound (CRB) [2] and also to compare the precision obtained with full and compact configurations [3]. In this communication, we analyze the robustness of the estimation proposed in [1] with respect to several RVoG deviations. The proposed robustness analysis is based on theoretical results obtained in [4], with the experimental setting of BIOMASS mission [5] and with vegetation characteristics observed in TropiSAR campaign data [6]. The BIOMASS context For the BIOMASS mission [4], the range resolution is limited by the small frequency allocation for remote sensing applications in the corresponding band. The authorized bandwidth is 6MHz which results in a 25m range resolution or a 59m ground range resolution at 25° incidence angle. The azimuth resolution is 12.5m. The expected level II products have a resolution of 0.25ha, 4 ha and 100ha. As a result, the number of looks is respectively, 3, 54 and 1352 for these three estimation window sizes. Procedure In this paper, we analyze the consequence of several model deviations on the performance of the estimator. The accuracy of an estimator can be described by the mean square error (MSE) where the MSE can be decomposed in a sum of two terms: the square of the bias and the variance. We use Monte Carlo simulations to estimate the bias and the variance, for the considered model deviations and for different estimation window sizes. In these simulations, the RADAR characteristics are based on the BIOMASS mission characteristics, and the other parameters are representative of what can be observed in TropiSAR data [5]. We consider separately several model deviations. We analyze the consequence of a varying value of the vegetation or ground height in the window. We also analyze the consequence of an erroneous a priori on the vegetation attenuation and on the vertical repartition of the scatterers. The bias describes the systematic error caused by the model deviation, whereas the variance describes the precision around the mean of the estimator and is due to the speckle. We will present on several examples and window sizes the variance of the considered estimator and the Cramer Rao bound (CRB) [3] for both compact and full polarizations. We will also use the expression linking the bias of the estimation to the standard deviation of the disturbances proposed in [6], where the expression is the same for full and compact modes. Analysis The analysis is performed on the TropiSAR dataset, over the Paracou site in French Guiana. Several assumptions underlying the RVoG model are considered and deviations from the model is are explored. Preliminary analysis lead to the following observations on the considered examples: • The attenuation does not seem dependent on the polarization and the bias on the vegetation height introduced by a 0.1dB/m error on the attenuation is of the order of 1m. • Even in the extreme case of an elevated volume of 50% of the total vegetation height, the vegetation height canopy estimation is accurate. • The height of ambiguity impacts the equivalent number of looks (which thus impact the variance of the estimator) in the window because of the frequency shift but is also a key parameter in the sensitivity analysis to model deviation [4] (which thus impact the bias of the estimator). Trying to find a tradeoff between these two effects will thus complicate the choice of the baseline. • The spatial variation of vegetation height induces a bias, which can be estimated as proportional to the vegetation height standard deviation. For instance, with a standard deviation of 4 meters on the vegetation height in the estimation window, the bias is approximately around 1 meter for the analyzed examples. • The spatial variation of topography, if not accounted for, is a very critical parameter and can introduce a large bias in both ground and vegetation height estimates. For the analyzed examples representative of physical situations observed in real data, the bias on vegetation height can reach 4 meters. • The loss of performance of compact polarization compared to full polarization can be compared [3] to a loss of number of looks by a factor between 1.5 and 5 on the considered examples. This point is especially critical for BIOMASS. Acknowledgement: the authors wish to thank the conseil régional Provence Alpes Côte d’Azur for taking part in the funding of the PhD of Arnaubec and TOSCA, CNES for partial fundings of this research. References: [1] S. R. Cloude and K. Papathanassiou, Three-stage inversion process for polarimetric SAR interferometry, Institution of Electrical Engineers Proceedings, Radar, Sonar, and Navigation, 2003. [2] A. Arnaubec, A. Roueff, P.C Dubois-Fernandez and Ph. Refregier, Vegetation height estimation precision with compact PolInSAR and Homogenous random volume over ground model, in IEEE Transactions on Geoscience and Remote Sensing, vol. 52, num 3, pp. 1879-1891, Feb, 2014. [3] A. Roueff A. Arnaubec, P.C. Dubois-Fernandez and Ph. Refregier, Cramer-Rao lower bound analysis of vegetation height estimation with random volume over ground model and polarimetrix SAR interferometry, Vol 8, Issue 6, pp 1115-1119 in IEEE Geoscience and Remote Sensing Letters, 2011. [4] A. Arnaubec Analyse de la précision d’estimation de paramètres de la biomasse végétale à l’aide de techniques PolInSAR, Doctorat de l’université Aix Marseille III soutenu en octobre 2012. [5] T. Le Toan, S. Quegan, M.W.J. Davidson, H. Balzter, P. Paillou, K. Papathanassiou, S. Plummer, F. Rocca, S. Saatchi, H. Shugart, L. Ulander, The BIOMASS mission: Mapping global forest biomass to better understand the terrestrial carbon cycle. Remote Sensing of Environment, 115, 2850-2860, 2011. [6] The TropiSAR Airborne Campaign in French Guiana: Objectives, Description, and Observed Temporal Behavior of the Backscattered Signal, P-C. Dubois-Fernandez, T. Le Toan, S. Daniel, H. Oriot, J. Chave, L. Blanc L. Villard, M-W-J. Davidson and M. Petit. IEEE Transactions on Geosciences and Remote Sensing, Vol 50, Issue 8, pp 3228-3241, 2012.

  
   

• 11:50 On the Characterisation of Scattering Mechanisms Occurring in Agricultural Vegetation by Means of Polarimetric SAR Tomography

  Joerg, Hannah (1,2); Pardini, Matteo (1); Hajnsek, Irena (1,2) 1: German Aerospace Center, Germany; 2: ETH Zürich, Institute of Environmental Engineering

  Show abstract

  The interpretation of scattering mechanisms occurring in agricultural vegetation at different frequencies is rather difficult, due to the increased complexity (compared to forest) of the scattering processes and at the same time the limited observation space provided by conventional, even multi-parameter, SAR configurations. Indeed, the vegetation volume might be oriented and most importantly, the structure of the plants can change very fast over the phenological cycle as well as from one crop type to the other [1,2]. (Polarimetric) SAR Tomography enables to resolve the scattering processes in the vertical dimension [3] and therefore to assess the differences in the 3-D scattering distribution with respect to species and development stage at different polarisations or frequencies. However, there is a lack of multi-baseline, multi-frequency polarimetric SAR datasets over agricultural areas with a sufficient height resolution and/or a proper temporal sampling (covering the phenological cycle) [1,4]. In order to fill this gap, a new campaign was conducted in 2014 with DLR’s airborne sensor F-SAR over an agricultural area near Wallerfing, Germany. Fully polarimetric multi-baseline SAR acquisitions in X-, C- and L-band were acquired almost weekly from May to August covering different moments of the phenological cycle. The baselines were planned to be big enough to ensure a high vertical resolution required to image low vegetation heights and in a sufficient number to allow accurate inversion. More in detail, up to 9 parallel tracks have been acquired, per frequency and acquisition date, allowing a vertical resolution (in mid range) down to 0.5m at X-band, 0.8m at C-Band and 1.4m at L-band and a height of ambiguity bigger than 9m at all frequencies. In parallel, ground measurements of soil and vegetation parameters were taken on each acquisition date. The data are processed using non-model based as well as model based spectral estimators [3,5]. In addition, the polarimetric information is explored to characterise the occurring scattering mechanisms by comparing the different channels and by analysing jointly the whole polarimetric 3-D signature [6]. In this work, first results obtained from this unique data set to interpret the temporal and spatial behaviour of polarimetric scattering mechanisms occurring in agricultural vegetation are presented and discussed. First, the scattering processes along the phenological cycle are investigated by comparing different growth stages of a certain crop type under study. Then, moving from higher (X-band) to lower (L-band) frequencies, the frequency dependency of ongoing scattering mechanisms is addressed in terms of differences in penetration, ground contribution and the amount of volume scattering. Finally, analysis of the polarimetric 3-D scattering distribution through the phenological cycle enables to assess the validation of coherent scattering models used to interpret and to invert polarimetric interferometric SAR data. [1] S.C.M. Brown, S. Quegan, K. Morrison, J.C. Bennett, G. Cookmartin: High-Resolution Measurements of Scattering in Wheat Canopies – Implications for Crop Parameter Retrieval, IEEE Transactions on Geoscience and Remote Sensing, Vol. 41, no. 7, pp. 1602-1610, 2003. [2] D. Ballester-Berman, J.M. Lopez-Sanchez, J. Fortuny-Guasch: Retrieval of Biophysical Parameters of Agricultural Crops Using Polarimetric SAR Interferometry, IEEE Transactions on Geoscience and Remote Sensing, Vol. 43, no. 4, pp. 683-694, 2005. [3] A. Reigber, A. Moreira: First demonstration of airborne SAR Tomography using multibaseline L-band data, IEEE Transactions on Geoscience and Remote Sensing, vol. 38, no. 5, pp. 2142-2152, 2000. [4] H. Jörg, M. Pardini, I. Hajnsek, K.P. Papathanassiou, First Multi-Frequency Investigation of SAR Tomography for Vertical Structure of Agricultural Crops, Proceedings of EUSAR, Berlin, 2014. [5] S. Tebaldini, Single and multipolarimetric SAR Tomography of forested areas: A parametric approach, IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 5, pp. 2375-2387, 2010. [6] L. Ferro-Famil, Y. Huang, A. Reigber , High-Resolution SAR Tomography Using Full Rank Polarimetric Spectral Estimators, Proceedings of International Geoscience and Remote Sensing Symposium (IGARSS), Munich, 2012

  
   

• 11:30 TanDEM-X Pol-InSAR Inversion for Mangroves of East Africa

  Lee, SeungKuk; Fatoyinbo, Temilola; Lagomasino, David; Osmanoglu, Batuhan NASA Goddard Space Flight Center, United States of America

  Show abstract

  Mangrove forests are found along the intertidal zones of tropical and subtropical regions where wave energy is sufficiently low and suitable sediments are present. These coastal forests represent only 3% of the total forested areas of the world, but they are amongst the highest carbon-storing and carbon-exporting ecosystems, globally. Carbon measurements in mangrove forests can be inferred by using effective passive and active remote sensing methods. Passive methods include using optical remote sensing techniques to estimate the horizontal structure (e.g., extent and distribution) of mangroves across the globe, while, global InSAR and lidar data sets have become available to calculate the forests’ vertical structure (e.g., height and biomass). However, those datasets have had relatively lower spatial and temporal resolutions when compared to conventional SAR satellite systems. TanDEM-X (TDX) is an innovative satellite mission that is flying twin X-band SAR satellite sensors in order to generate a consistent global DEM with high resolution and accuracy using a bistatic mode. One satellite acts as a transmitter, while both satellites simultaneously receive the returned signal. The TDX has additionally given us the first acquisition of satellite Pol-InSAR data without a temporal decorrelation effect. However, using the TDX data for forest Pol-InSAR applications, the penetration capability of X-band into the volume and the polarimetric diversity of interferometric coherence become an issue. In addition, in the case of single-pol TDX acquisition, inversion relies on external information because of the underdetermined problem in Pol-InSAR. To solve the problems related to TDX Pol-InSAR inversion, the suggested ground phase in the Pol-InSAR model was estimated directly from TDX interferograms with the assumption that the underlying topography in mangrove forests is flat (i.e. water surface). Single- and dual-pol inversion approaches were applied to estimate mangrove tree height from TDX data sets without any external underlying topographic information. In the present study, we proposed to generate 12m-resolution mangrove canopy height and biomass estimates for the coastline of East Africa using Pol-InSAR techniques from single-/dual-pol TDX data. The TDX height data were calibrated with commercial airborne lidar data acquired over 150 km2 of mangroves in the Zambezi Delta of Mozambique while height and biomass estimates were validated using in-situ forest inventory measurements. The results from the present study are the first country-wide, wall-to-wall estimates of mangrove structure and biomass at 12 m resolution.

  
   

• 12:10 Forest height estimation using Space-borne PolInSAR dataset over Tropical forests of India

  Khati, Unmesh Govind; Kumar, Shashi; Agrawal, Shefali Indian Institute of Remote Sensing, ISRO, Dehradun, India

  Show abstract

  Polarimetric SAR Interferometry has gained prominence due to its ability of combining the advantages of the two well established techniques of SAR Polarimetry and Interferometry. Many studies have analyzed airborne PolInSAR data for extraction of target scattering and physical properties. With the launch of fully polarimetric space-borne SAR systems such as the Canadian RADARSAT-2 and Japanese ALOS PALSAR, capable of acquiring PolInSAR data, the technique can be applied for areas where airborne data is not available. Use of space-borne PolInSAR data is limited due to poor spatial and temporal resolution, and higher SNR as compared to airborne data. Much literature is not available for forest parameter estimation using space-borne PolInSAR data. One of the major application of the PolInSAR technique is estimation of forest parameters such as tree height and forest biomass. These parameters are important to determine the maturity of the forest, its ability to act as carbon sink and its carbon content. This paper demonstrates the use of space-borne PolInSAR data for tree height estimation in tropical forests of Uttarakhand State, India. C-band PolInSAR data is acquired using RADARSAT-2 platform over the tropical forest ranges of Barkot and Thano near Rishikesh City, Uttarakhand. The Barkot and Thano forests are situated between the Shivalik and Garhwal ranges of Himalayas and Barkot forest range is bordered to its east by The Ganges. The forest range is dominated by Sal (Shorea robusta) trees with the other dominating species being Teak (Tectona grandis), Sisoo (Dalbergia sisoo) and Khair (Senegalia catechu). The forest ranges cover a total area of around 150 sq.km. The spatial resolution of the multi-looked product is 12.5 meters and the temporal resolution is 24 days. The average coherence over the forested region ranges from 0.4 to 0.7. The major source of loss of coherence is temporal decorrelation during the gap of 24 days between acquisitions. The present work analyses two techniques for forest height estimation: Coherence Amplitude Inversion (CAI) technique and Three Stage Inversion (TSI) technique. In CAI, two polarization channels are selected, one dominated by surface scattering and the other by volume scattering. It calculates the tree height using estimates for extinction in the volume layer. The present work also investigates the TSI technique developed by Cloude and Papathanassiou. The TSI technique uses the two-layer model for vegetation. TSI technique observes the complex coherence in different polarizations and predicts the ground topography and polarization-independent volume coherence using least squares approach. It considers the effects of temporal decorrelation on estimation of height of the vegetation layer; it also utilizes the phase information and estimates vegetation structure through use of wave extinction. The basic observable in PolInSAR data is the 6x6 coherence T6 matrix. Using polarization basis transformations, complex coherence is calculated in different polarization basis such as Linear, Circular, Pauli and Optimal basis. It is shown that the complex coherences are a function of ground-to-volume scattering ratio. It is also shown that among all the channels the Pauli basis channels, HV+VH best represents the volume scattering and HH+VV channel best represents the surface or ground scattering. This knowledge is used for selection of optimum channels for forest height estimation using CAI technique. Validation of obtained results is carried out using field collected tree height. 100 plots were plotted spread through-out the forest areas of Barkot and Thano forests. Square plots of 12.5 meter side length are established in the forest range and the parameters such as maximum tree height per plot, average tree height per plot and dominant species in the plot area are noted. These parameters are later used for validation of the obtained results. The field measured forest height ranges from a minimum of 15 meters for Khair-Sisoo forested regions to a maximum of 29 meters for dense Sal and Teak forest. The mean forest height is 23.44 meters with a standard deviation of 2.87 meters. The CAI technique is applied over the study area for forest height estimation. The HH+VV and HV+VH polarization combinations are selected representing ground and volume scattering channels. The estimated forest height using CAI technique ranges between 20m – 27m with a mean of 23.61m. When compared with the field measured height at 100 locations, the CAI technique presents a correlation of 0.34 and RMSE of 2.77m. The second technique for forest height estimation is the TSI technique. The forest height estimated using the TSI technique ranges between 18m and 27m with a mean of 22.98m. The correlation between the field height and the estimated height of vegetation is 0.62. The average accuracy of the CAI and TSI method is 90.10% and 91.56% respectively. It is observed that the forest height is underestimated in both the techniques. Probable reasons for underestimation of forest height are discussed and plot locations are identified which lead to high source of error in forest height estimation. After eliminating the 16 identified plots, the correlation between actual and estimated forest height increases to 0.712 for TSI technique with an average accuracy of 94.07%. The present work signifies the importance of PolInSAR based tree height estimation of tropical forests. The two PolInSAR techniques are compared, and it is concluded that the ‘Three Stage Inversion’ technique is a robust and accurate technique for forest height estimation. This method is suitable for measuring the height of large swaths of forested areas. In the present study, the technique is applied to tropical forest area, it may also be applied to evergreen inaccessible forested areas. A major limitation for validation of this study is the non-availability of airborne LIDAR data for tree height validation. With LIDAR data, the tree height can be validated in larger areas which are inaccessible for conducting ground surveys. Keywords: PolInSAR, height, space-borne, Radarsat-2, polarimetric interferometry, SAR

  
   

Highlights

In May 2013, the Biomass mission concept was selected to become the next in the series of satellites developed to further our understanding of Earth.

Released: 21/06/2013

REPLAY EARTH EXPLORER 7 USER CONSULTATION MEETING (PART 1)

Released: 11/03/2013

POLinSAR 2013 Workshop

Released: 22/02/2013