Forest Vertical Parameter Estimation Using PolInSAR Imagery Based on Radiometric Correction

Transcription

1 International Journal of Geo-Information Article Forest Vertical Parameter Estimation Using PolInSAR Imagery Based on Radiometric Correction Yu Zhang 1, Chu He 1,2, *, Xin Xu 1 and Dong Chen 3 1 Signal Processing Lab, Electronic and Information School, Wuhan University, Wuhan 4379, China; yuzhang_spl@whu.edu.cn (Y.Z.); xinxu@whu.edu.cn (X.X.) 2 State Key Laboratory for Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 4372, China 3 ATR Key Lab, National University of Defense Technology, Changsha 4173, China; dongcheny@sina.com * Correspondence: chuhe@whu.edu.cn; Tel.: Academic Editor: Wolfgang Kainz Received: 3 June 216; Accepted: 28 September 216; Published: 1 October 216 Abstract: This work introduces an innovative radiometric terrain correction algorithm using PolInSAR imagery for improving forest vertical structure parameter estimation. The variance of radar backscattering caused by terrain undulation has been considered in this research by exploiting an iteration optimization procedure to improve backscattering estimation for a Syntic Aperture Radar (SAR) image. To eliminate variance of backscatter coefficients caused by local incident angle, a radiometric normalization algorithm has been investigated to compensate influence of terrain on backscattering values, which hinders forest vertical parameter estimation. In vertical parameter estimation, species diversity and spatial distribution of different vegetation have been modeled. Then, a combination of Fisher s Alpha-Diversity model parameter estimation and three-stage inversion method was designed for vertical structure parameter. To demonstrate efficiency of proposed method in forest height estimation, classical phase difference and three-stage inversion approach have been performed for purpose of comparison. The proposed algorithm is tested on ALOS PALSAR (Advanced Land Observing Satellite Phased Array type L-band Syntic Aperture Radar) and RADARSAT-2 (Radio Direction and Range Satellite 2) data sets for Great Xing an Mountain area and BioSAR (Biomass Syntic Aperture Radar) 27 data sets for Remningstorp area. Height estimation results have also been validated using in-situ measurements. Experiments indicate proposed method has ability to compensate influence of terrain undulation and improving accuracy of forest vertical structure parameter estimation. Keywords: PolInSAR; radiometric terrain correction; vertical structure; scattering mechanism; RVoG model (Random Volume-over-Ground model) 1. Introduction Forest parameters, in particular vertical structure parameters, are prerequisite of dynamic analysis of carbon cycle and water cycle [1,2], which play an important role in research regarding environmental protection and global climate system [3]. Forest height is one of most essential parameters of forest information. However, terrain variance poses a challenge for parameter inversion based on PolInSAR imagery because undulation may make vegetation backscatters more complex. Polarimetric Interferometry SAR (PolInSAR) remote sensing technology has potential in forest parameter estimation since it has sufficient penetration ability to extract ground and canopy information of forest. Large-scale and long time series observation of scene can be provided to model dynamical information of forest area. Forest height estimation is a hot topic in PolInSAR remote sensing. ISPRS Int. J. Geo-Inf. 216, 5, 186; doi:1.339/ijgi

2 ISPRS Int. J. Geo-Inf. 216, 5, of 21 The phase difference method and model based method algorithm are main inversion methods. The phase difference method utilizes interferometry phase to represent canopy and corresponding ground for inversion of forest height. Based on interferograms of different scattering mechanisms obtained from coherence optimization method, Cloude [4] first uses PolInSAR technology to estimate forest height. Gabriel [5] proposes coherence decomposition approach, where polarimetric decomposition is first used to classify different scattering mechanisms and n a phase difference procedure involving ground phase and canopy phase is conducted to obtain forest height. Yamada [6] considers spectral analysis ory of super-resolution technology, as well as ESPRIT approach to estimate orientation of signal waves, which represents first attempt to incorporate super-resolution technology into research on vegetation height inversion. To extract interference phase in situations characterized by limited observation data, Wang [7] develops ESPRIT method to calculate interference phase and scattering power. Scattering power was utilized to classify scattering mechanism, which uses interferograms to represent physical characteristics of real landscape scene. Guilaso [8] suggests that ESPRIT does not exhibit accuracy in forest terrain phase. The simulation experiment indicates that distortion of terrain may cause unreliable estimation in ground scattering phase, retain depolarization components and lead to Ground-over-Volume ratio of less than 3 db. The model based method depends on complex coherent coefficient model of PolInSAR data, and has a strong physical and oretical basis. In addition to height of vegetation, extinction coefficient and or vertical parameters can be derived. Based on PolInSAR complex coherent coefficient model proposed by Oveisgharan [9], Papathanassiou transforms inversion procedure into a classical non-liner parameter estimation problem, which is referred to as six-dimension non-liner iteration method. It provides many valuable parameters simultaneously, however, calculation is time consuming, and it strongly depends on initial value and tends to fall into local optimal solution [1]. To solve se problems, methods such as simulated annealing method [11], genetic algorithm [12] and BP neural network [13] are proposed. Cloude [14] proposes three-stage inversion method under assumption that Ground-over-Volume ratio of different channels is located on a straight line and estimation accuracy of volume coherence coefficient is less than 1 db. Thus, height estimation error can be less than 1%. Moreover, Cloude [15] proposed a multi-baseline and multi-frequency PolInSAR based method that aims at extracting ground and canopy phase in redundant interferometric channel to improve accuracy of height estimation result. However, calculating so many parameters for optimal interferometric phase extraction is time consuming. The variance in terrain has a strong effect on vegetation height estimation. In rugged areas, terrain undulation may lead to substantial differences in ground echo, reby changing SAR image backscatters. As for geometric aspect, terrain distortion changed location of image pixel. It may lead to shadow, layover and foreshortening in SAR imaging area. In terms of radiometric aspect, radar backscattering relies on terrain since corresponding local incident angle could change backscattering coefficient and scattering area due to terrain variance. Thus, land cover category will be different. As a result, even for same land cover categories, radiometric values may differ, while different land cover categories may have a similar brightness value, which makes vegetation parameter inversion challenging work. To eliminate distortion in SAR imaging area, terrain correction method is introduced via elevation information provided by Digital Elevation Model (DEM). The typical method includes geometric correction and radiometric correction. Geometric correction aims at locating each pixel in right position, in particular real location of ridges and valleys. In terms of geometric correction methods, Polynomial method [16] is a traditional approach that simulates deformation area while ignoring geometric procedure in image area. Computation is fast but only suitable in situation of relatively flat terrain. Based on imaging ory of SAR, Leberal [17] proposes slant distance function and Zero Doppler function

3 ISPRS Int. J. Geo-Inf. 216, 5, of 21 for airborne SAR images. This algorithm is insufficient for space borne SAR images since Doppler effect is not zero and distance migration is relatively significant. The DEM based simulation method first constructs simulating image using SAR imaging parameter and DEM information for simulation image calibration to build coordinate function between SAR image and DEM. Schreier [18] introduces concept of Geocoding Ellipsoid Corrected (GEC), which transforms orbit model, range function and Doppler frequency function into a parameterized polynomial to build Range Doppler (RD) model. The resampling-based iteration method has also been used to obtain real location of SAR imagery pixel. Meier [19] introduces Geocoded Terrain Corrected (GTC) method based on interpolation of satellite location and speed vector without parameterized orbit function. In correction step, Doppler function-based iteration procedure is used to calculate image coordinate from ground coordinate without ground control points (GCP). Radiometric correction is performed by correcting local terrain brightness, which is defined by incident angel, local terrain gradient and azimuth. To transform radar brightness into backscattering coefficient, local terrain and imaging geometric should be considered. Bayer and Mauser [2 22] demonstrate that radar brightness can be affected by terrain in two respects. First, scattering area may be different for same unit in different terrain conditions, and same ground cover may have different scattering mechanisms because of different local incident angles induced by different terrains. For same scattering unit in slant range, forward scattering area is large, and backward area is small in ground range due to terrain gradient. Small [23,24] takes full consideration of geometric relationship between SAR imaging and DEM information in proposed homogeneity and heterogeneity ory. As for homogeneity, radiometric normalization method is used when corresponding pixel in DEM has no effects on SAR image. As for heterogeneity, geometric relationship between DEM and SAR image should be considered. Each DEM pixel area consists of several SAR pixels; n, DEM pixel can be divided into several units, and effective scattering area is represented as a combination of several units. Lars [25] proposes a project angle method for terrain correction result in map space, which can significantly reduce computation cost. The different incident angles could result in huge variance in land cover backscattering. Geometric correction methods, such as DEM simulation and RD resampling method based on interpolation, fail to achieve precision in forest parameter estimation. The radiometric correction method depends on slope, and terrain is relatively flat while current method is ineffective under fractured terrain conditions. The traditional inversion procedure may limit accuracy of vertical structure parameter estimation. Current radiometric terrain correction methods mainly focus on eliminating influence of amplitude of backscatter, while phase of backscattering is ignored. In this paper, we proposed an improved terrain correction and height estimation algorithm by considering tree species diversity and terrain correction. Different tree species have different vertical structures and spatial distributions, and effect of incident angle on different species is different. Land cover characteristics in SAR images are analyzed using a scattering mechanism extraction procedure via land cover classification of forest area. In terrain radiometric correction, an iteration optimization procedure is investigated for optimal backscattering estimation to compensate influence of terrain distortion. In vertical structure parameter extraction procedure, species diversity and spatial distribution of vegetation are modeled. The vertical structure parameter is retrieved based on interferometric coherence coefficient model. Compared with direct three-stage inversion, estimation accuracy of forest height is improved. This paper is organized as follows. Section 2 describes basic characteristics of a SAR image in forest area. A detailed framework of proposed algorithm is presented in Section 3 after a brief introduction to scattering mechanism classification-based radiometric correction algorithm and vertical parameter estimation method. In Section 4, experiments to verify proposed method and comparison methods from terrain correction analysis and tree height estimation are conducted on real ALOS PALSAR, RADARSAT-2 and BioSAR data sets. Finally, discussions and conclusions are presented in Section 5.

4 ISPRS Int. J. Geo-Inf. 216, 5, of PolInSAR Characteristics Analysis in Forest Area PolInSAR image processing is a potential technique for terrain correction and vertical structure parameter extraction. In this section, PolInSAR characteristics are analyzed in terms of polarimetric SAR interferometry and backscattering PolInSAR Character Analysis The PolInSAR data is acquired by two slightly different positions in space of two fully polarimetric SAR images. In monostatic situation, each PolSAR image can be expressed as Pauli target components [26,27] K i = 1 2 [S hh,i + S vv,i, S hh,i S vv,i, 2S hv,i ] T (1) Different polarization states represented by transmitting and receiving electromagnetic waves are combined to generate complex SAR data with a horizontal h or vertical v polarization basis, where i = 1, 2 indicates each one of interferometric data sets. In Pauli vector condition of Equation (1), PolInSAR target vector is constructed from combination of K i, for i = 1, 2. K i = [ K T 1, KT 2 ] T (2) For deterministic or point scatters, such as dihedral scattering, Equation (2) is a deterministic vector. As for distributed scatters, Equation (2) is a random vector as a consequence of complexity of scattering process [26]. For stationary data, Equation (2) can be represented as a six-dimensional, zero-mean, complex Gaussian probability density function (pdf). Under this assumption, useful PolInSAR information can be transformed into a second-order moment as a coherency matrix. It can be reformulated as: ] [ { T 6 = E KK H} = T 11 Ω 12 Ω12 H T 22 where H indicates complex transposition and E{ } is expectation operator. The matrices T 11 and T 22 correspond to individual polarimetric coherency matrices, and Ω 12 is polarimetric interferometric coherency matrix [4]. Then, interferometric complex correlation coefficient of two complex SAR images S 1 and S 2 is obtained as Equation (4): ρ = ρ e jφ x = E (3) E {S 1 S2 } { S 1 2} E { S 2 2} (4) with ρ 1. In bare region, phase is related to topographic height. The amplitude ρ, also known as interferometric coherence, determines accuracy of φ x Backscatter Conventions In SAR imaging areas, ratio between scattering power P s and incident power P i at ground level is defined as radar backscatter β. When target is a distributed target, each reference area gets backscatter ratio from backscatter coefficient. Generally, re are three reference areas (see 1). The solid rectangle A β in slant range as reference area and beta naught β can be expressed as [28] β = β/a β (5) When reference area is locally tangent to an ellipsoidal model of ground surface A σ in ground area, result is sigma naught σ E. σ E = β Aβ A σ = β sinθ E (6)

5 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, of 21 A E sine (6) A When reference area is defined to be in plane perpendicular to line of sight from sensor When to an ellipsoidal reference model area is ofdefined ground to be surface in plane A γ, gamma perpendicular naught γto E is line result. of sight from sensor to an ellipsoidal model of ground surface Aγ, gamma naught γ E is result. γ E = β Aβ A = β tanθ E (7) E A γ tane (7) A The σσ E or γ E backscatter values can be terrain-geocoded using digital height E. or γ E backscatter values can be terrain-geocoded using a digital height model (DEM), i.e., i.e., resampled resampled into into a map a geometry, map geometry, producing producing a geocoded-terrain-corrected a geocoded-terrain-corrected (GTC) product (GTC) [29]. Although product [29]. position Although or geometry position of or geometry backscatter of estimate backscatter has beenestimate correctedhas in GTC been products, corrected in radiometry GTC products, of resulting radiometry image of remains resulting ellipsoid-model image remains based. ellipsoid-model based. S E LIM E T A p A a g A A A A Terrain Terrain of of vegetation vegetation area area in SAR in SAR image: image: SAR imaging SAR imaging geometry; geometry; and SAR and backscattering. SAR backscattering. 3. Scattering Mechanism Classification Based Radiometric Terrain Correction Method for Height 3. Scattering Mechanism Classification Based Radiometric Terrain Correction Method for Height Inversion Inversion with with Species Species Diversity Diversity Coefficient Coefficient Estimation Estimation 3.1. Iteration Based Backscattering Coefficient Optimization To eliminate effects of of terrain induced geometric and and radiometric distortion distortion and and maximize maximize correlation correlation with with biophysical biophysical parameter parameter as well as well minimize as minimize root mean root square mean error square of vegetation error of parameter vegetation estimation, parameter aestimation, scattering a mechanism scattering classification-based mechanism classification-based radiometric terrain radiometric correction terrain is introduced correction is to improve introduced to accuracy improve of backscattering accuracy estimation. of backscattering Let polarimetric estimation. covariance Let polarimetric matrices Ccovariance matrices C i C P P i C P P be given for i [1,.. be., n], given where for P iis [1, dimensionality, n], where P ofis polarimetry. dimensionality In case of of polarimetry. conventional In polarimetric case of conventional SAR measurements, polarimetric backscattering SAR measurements, at an arbitrary backscattering combination an of transmit arbitrary and combination receive polarization of transmit isand given receive by: polarization is given by: i (8) σ i = ω C i ω (8) where ω C P is a singular weighting vector ω ω and represents Hermitian transpose. For where quad-pol ω polarization C P is a singular basis P = weighting 4, ω can be vector represented ω ω and by represents electromagnetic Hermitian wave polarization transpose. For orientation quad-poland polarization ellipticity basis angles P = 4, associated ω can be represented with sphere, by electromagnetic Ψ and Χ for wave receiving polarization and orientation transmitting and polarizations ellipticity angles r, t. associated with sphere, Ψ and X for receiving and transmitting polarizations r, t. pr1 pt1 p r1 p r1 t1 t2 r, r, t, t p (9) ω (Ψ r, χ r, Ψ t, χ t ) = r1 p t2 r2 t1 (9) p r2 p t1 pr2 pt2 P t/r,1/2 are first and second elements of unitary wave polarization vectors for receive and transmit polarizations, respectively p r2 p t2

6 ISPRS Int. J. Geo-Inf. 216, 5, of 21 P t/r,1/2 are first and second elements of unitary wave polarization vectors for receive and transmit polarizations, respectively ˆp r/t (Ψ, χ) = [ cosψ sinψ ] [ sinψ cosψ cosχ isinχ In this case, ω can be modeled as a selection of scattering mechanism type, including isotropy versus anisotropy, even versus odd bounce, and horizontally versus vertically oriented. Thus, selection of polarization affects sensitivity to a specific scattering mechanism in illuminated resolution cell, which has a strong correlation with terrain. The polarization basis of C i can vary simultaneously by applying singular transformation matrix U C P P, which can be rewritten as: C i = UC i U (11) To eliminate distortion induced by terrain, key problem is to find polarization state that is most sensitive to effect of vegetation, which can be formulated by: (1) maximizing absolute value of correlation (R) between backscattering and prior polarization information with biophysical parameter from tie points of original measurements; (2) maximizing coefficient of sensitivity factor between specific scattering mechanism and backscattering (R 2 ); and (3) minimizing root mean squared error of inverted estimates (RMSE). If relationship between backscattering coefficient and polarimetric from prior information is linear, optimal procedures to maximize R and R 2 and minimize RMSE for estimating weight parameter ω are unbiased, and all three optimal problems are equivalent. The correlation between polarization information y i, which indicates relationship between land cover and corresponding backscattering in SAR image, and backscatter coefficient σ i(ω) is described as R (ω) = ( ) (y i y) σ i(ω) σ (y i y) 2 ] (1) ( ) (12) 2 σ i(ω) σ Thus, polarization-dependent optimization of absolute value of correlation can be formulated as (ω ω = argmax C i ωy i) (ω C i ωy i ) = argmax (ω C i ω 1 n (ω C i ω)) 2 (y i y) 2 ω (13) A i ω (ω B i ω) 2 where arg max () operation stands for argument of maximum and A = (C i y i ) nĉy, B i = C i Ĉ, Ĉ = 1 n C i. All C i, Ĉ, A are Hermitian positive semi-definite, while B i is Hermitian. To solve optimization problem of Equation (13), an iterative procedure is used to obtain approximate solution. The principle of optimization procedure is to transform C i to optimal polarization and optimize an approximate correlation function in new polarization basis, using inequality Equation (14): n n (ω B i ω) 2 ω B i B iω (14) i=1 i=1 Then, optimization function in Equation (13) can be represented as below: ˆω argmax ω ω ω Aω n i=1 B i B iω argmax ω ω Aω ω Fω (15)

7 ISPRS Int. J. Geo-Inf. 216, 5, of 21 Equation (15) is used to maximize lower boundary of function, where F is a unique Hermitian positive semi-definite square root of B i B i. To simplify calculation, we neglect absolute value operation in Equation (15). Then Lagrangian multipliers can be used to solve optimal problem as following: L = ω Aω λω Fω (16) We obtain optimization result by maximizing numerator and keeping denominator as a constant of Equation (15) simultaneously. If sensitivity ω Aω is negative, it may lead to optimal correlation being less than zero. In this case, matrix A should be multiplied by 1 to find optimal polarization. With respect to derivation of L in Equation (16), solution can be transformed into an Eigen-problem. F 1 Aω = λω (17) The problem formulation and solution in Equations (16) and (17) provide an initial approximation, which is close to optimum. To improve result furr, an iterative approach is investigated. It is based on alternatively optimizing L in Equation (16) and transforming polarization basis of C i toward a state with optimal polarization, as shown in Equation (8). As a result, we can optimize estimation of backscattering coefficient by solving problems that are sensitive to terrain induced distortion Radiometric Based Terrain Correction Terrain undulation induced by local incident angle in PolInSAR imagery leads to a decrease in interferometric, which may result in an unstable estimation of height in forest area. To solve this problem, radiometric based terrain correction is proposed. The procedure of proposed method is demonstrated using following steps: 1. Pre-processing: A radiometric calibration step is used to convert original SAR image into radar brightness β, and multi-look and filtering process is conducted to obtain slant SAR data. 2. DEM area calculation: Due to fact that edge points of DEM have different elevations, projection area on ground range corresponding to DEM grid can be represented by summation of two triangles in 2a for accurate projection area calculation. In experiment, we choose ASTER DEM with a resolution of 3 m, downloaded from ersdac.jspacesystems.or.jp/. 3. LUT (Look Up Table) and projection calculation on ground range: First, this procedure calculates original SAR image coordinate (r init, a init ) and local incident angle corresponding to DEM cell. Then, projections on γ plane dγ DEM (N, E) are achieved based on local incident angle. Finally, SAR coordinate and local incident angle are combined to build LUT. 4. SAR image simulation: The original simulation result of SAR image can be obtained by using LUT and projections on γ plane. Since coordinates in LUT were not integers, area should be assigned to neighbor pixel cell. Then, it can be realized through a bilinear interpolation procedure shown in 2b, where (int) means integer operation. dσ DEM (N, E) = S T T 1 T 1 + S T11 T 1 T 12 (18) A γ (r, a) = [dσ DEM (N, E)] (19) (N,E) A 5. Registration of simulation and real SAR image: We first calculate offset between original SAR image and simulation SAR image to build a registration polynomial. Then, a cross searching procedure is performed to obtain match point. Next, match point is used to build quadratic polynomial function to revise LUT. Finally, we can get modified simulation SAR image from γ projection and revised LUT.

8 ISPRS Int. J. Geo-Inf. 216, 5, of Radiometric correction: The radiometric correction result on slant range can be obtained by area integral on γ plane. Then, Orthophoto of final terrain correction result is obtained by using revised LUT through a bilinear interpolation procedure. σt = β A β cosθ (2) ISPRS Int. J. Geo-Inf. 216, 5, 186 A γ 8 of 21 T 1 T 1 ((int)r, (int)a) ((int)r+1, (int)a) T T 11 (r,a) E 1 E 1 E E 11 ((int)r, (int)a+1) ((int)r+1, (int)a+1) 2. DEM 2. DEM projection: projection: DEM DEM area calculation; area calculation; and and bilinear bilinear interpolation. interpolation., DEM, 3.3. RVoG Model Based Forest Vertical A r Structure a Modeling d N E (19) N, EA For PolInSAR based vegetation height estimation, phase difference method failed to 5. extract Registration real of simulation phase center and ofreal SAR canopy image: andwe ground; first calculate it tends to offset underestimate between original real height. The SAR six-dimension-based image and simulation non-liner SAR image iteration to build approach a registration haspolynomial. a strong dependence Then, a cross onsearching original parameter procedure and is performed selection to of obtain polarization match channel. point. Next, Moreover, match it is often point time is used consuming to build and lacks aquadratic physical mechanism. polynomial function Thus, weto introduced revise LUT. improved Finally, RVoG we can model get based modified three-stage simulation inversion procedure SAR image tofrom estimate γ projection vegetation and height. revised LUT. 6. Radiometric The most correction: importantthe aspect radiometric of PolInSAR correction data analysis result on is slant dependency range can of be obtained interferometric by information area integral on polarization. on γ plane. To derive Then, underlying Orthophoto topography, of final terrain polarimetric correction effects result is of ground, obtained including by using direct revised andlut double-bounce through a bilinear effects, interpolation rely on procedure. principle that double-bounce effect of polarimetric phase has opposite signs A cos in different terms of polarimetric interferometric covariance matrix, while interferometric T phase appears with same sign [26]. (2) A Forest scattering is a two-layer model, shown in 3. One is volume scattering from canopy and or is surface scattering from ground. To acquire a better estimation of 3.3. RVoG scattering Model center Based offorest ground Vertical and Structure volumemodeling scattering, scattering mechanism classification method is For exploited. PolInSAR For based volume vegetation scattering, height it estimation, can be separated phase if difference projection method vector failed of to extract two fully real polarimetric phase center images of is properly canopy selected. and ground; it tends to underestimate real height. The six-dimension-based non-liner iteration approach ( has ) H a strong dependence on original parameter and selection of polarization channel. ωmoreover, 1 V T V 2 ω2 V it = is often time consuming and lacks a (21) physical mechanism. Thus, we introduced improved RVoG model based three-stage inversion procedure where to ω1 V estimate and ωv 2 mean vegetation projection height. vector and TV 2 represents volume scattering of image. To The separate most important surface aspect scattering of PolInSAR ˆT g, extended data analysis Freemen Durden is dependency [3] decomposition of interferometric is used. information on polarization. To derive underlying topography, polarimetric effects of ground, including direct and T 11 (1, double-bounce 1) n 1 effects, rely on principle that doublebounce effect of polarimetric T 11 (3, 3) T 11 (1, 2) ˆT g = phase T 11 has (2, opposite 1) signs T 11 (2, in 2) different T 11 (3, terms 3) of polarimetric (22) interferometric covariance matrix, while interferometric phase appears with same sign [26]. Forest scattering is a two-layer model, shown in 3. One is volume scattering from where T canopy and 11 (i, j) represent element of column i and row j of polarimetric coherency matrix T or is surface scattering from ground. To acquire a better estimation of 11. scattering center of ground and volume scattering, scattering mechanism classification method is exploited. For volume scattering, it can be separated if projection vector of two fully polarimetric images is properly selected. V H V V 1 T2 2 (21)

9 1 T11 1,1 T11 3,3 T11 1,2 n ˆ Tg T11 2,1 T11 2, 2 T11 3, 3 (22) ISPRS Int. J. Geo-Inf. 216, 5, of 21 where T11(i, j) represent element of column i and row j of polarimetric coherency matrix T11. zz - h v h v r h v Two-layer Two-layer coherence coherence RVoG RVoG model model illustration illustration for vegetated for vegetated land surface. land surface. The different The different colors of colors arrow of onarrow left on represent left represent different scattering different scattering mechanisms mechanisms and corresponding and corresponding phases are shown phases on are shown right side. on The right surface, side. double-bounce, The surface, double-bounce, and short understory and short cause understory similar interferometric cause similar phases interferometric at ground phases level. at The volume ground layer level. introduces The volume a phase layer offset introduces and an additional a phase offset decorrelation and an due additional to phase decorrelation variation. due to phase variation. After scattering mechanism classification procedure, complex interferometric coherence After scattering mechanism classification procedure, complex interferometric coherence for random volume over ground (RVoG) is derived as shown in Equations (23) and (24), which is a for a random volume over ground (RVoG) is derived as shown in Equations (23) and (24), which is three-component singular complex vector defining choice of polarization [4], is extinction a three-component singular complex vector defining choice of polarization [4], σ is extinction coefficient of medium, kz is vertical wave number of interferometer (following spectral coefficient of medium, k range filtering) and θ is incident z is vertical wave number of interferometer (following spectral angle. The angles φ1 and φ2 are phase centers of bottom range filtering) and θ is incident angle. The angles ϕ of layers 1 and 2, respectively. 1 and ϕ 2 are phase centers of bottom of layers 1 and 2, respectively. * T 12 * T γ = ω T Ω 12 ω ω T T 11 ωt 11 T 11 = V 2 hv G T I 11 1 V I + e( 2σh v/cosθ 1 e ) I I1 G 1 Ω 12 = e iφ 2 I2 V + V e iφ 1e ( 2σh V)cosθ i2 I2 i1 2hV I cos G2 G 12 e e e I2 I1 V = eiφ 1e ( 2σh V)cosθ hv e ( 2σz )cosθ T v dz (23) 1 2 h cos h (23) V i V V 2 z ' cos I1 G = I h 1 V e e e T ' vdz δ (z ) e ( 2σz )cosθ T g dz = T g I2 V = h G V 2 z ' cos I e( 2σh V)cosθ h V e ( 2σz )cosθ e ik zz T V dz 1 z ' e T ' gdz Tg ( 2 h cos h V V V 2 z ' cos γ ikz z ' I= 2 e iφ e γ V + m e (ω) ) 1 + m (ω) (1 e γ TV) V dz ' (24) Equation (24) indicates that complex coherence i m coefficient in complex plane should be located in a straight line; two endpoints e are V γe 1V 1 i and e m i respectively. Since m(ω) equals zero, (24) pure vegetation volume coherence coefficient is obtained, and it can be used to estimate tree height and Equation extinction (24) coefficient indicates that in three-stage complex inversion. coherence coefficient in complex plane should be located in a straight line; two endpoints are γe i and e i respectively. Since m(ω) equals zero, 3.4. Framework pure vegetation of Terrain volume Correction coherence and Species coefficient Diversity is obtained, Based Vegetation and it Vertical can be Structure used to estimate tree Inversion Algorithm height and extinction coefficient in three-stage inversion. The framework of proposed method is shown in 4. First, scattering mechanism classification 3.4. Framework procedure of Terrain iscorrection exploitedand based Species on Diversity PolInSAR Based data Vegetation for land cover Vertical scattering Structure classification. Then, Inversion Algorithm iteration-based backscattering optimal method is investigated to obtain backscattering coefficient which is most sensitive to terrain distortion. The details are described in Section 3.1. Then, radiometric terrain correction method is investigated to eliminate effect of terrain undulation described in Section 3.2. Finally, after Fisher s alpha-diversity estimation, RVoG model based three-stage inversion procedure is introduced to obtain vegetation vertical structure parameters.

10 classification. Then, iteration-based backscattering optimal method is investigated to obtain backscattering coefficient which is most sensitive to terrain distortion. The details are described in Section 3.1. Then, radiometric terrain correction method is investigated to eliminate effect of terrain undulation described in Section 3.2. Finally, after Fisher s alpha-diversity estimation, ISPRS RVoG Int. model J. Geo-Inf. based 216, 5, three-stage 186 inversion procedure is introduced to obtain vegetation vertical 1 of 21 structure parameters. 4. Framework of terrain correction and species diversity based vegetation vertical structure inversion. 4. Framework of terrain correction and species diversity based vegetation vertical structure inversion. Based on distribution character of RVoG complex coherence model, main procedure of three-stage inversion algorithm contains three steps: Based on distribution character of RVoG complex coherence model, main procedure of 1. Least squares straight line fitting: Based on complex coherence coefficient of different polarization three-stage inversion algorithm contains three steps: channels, least squares method is used to fit straight line. If re are only two polarization 1. Least channels, squares fitting straight process is line degenerated fitting: Based for calculating on complex straight coherence line of two coefficient points. of different 2. polarization Terrain phase channels, calculation: least corresponding squares method complex is usedcoherence to fit coefficient straight line. of If re terrain are only interferometry two polarization phase channels, depends on fitting process intersection is degenerated of unit for circle calculating and fitting line. straight It can line be of obtained two points. by calculating relative location between channel with largest volume scattering and channel with largest surface scattering. After optimal of backscatter 2. Terrain phase calculation: corresponding complex coherence coefficient of terrain coefficient, terrain correction result is calculated with gamma naught and backscattering interferometry phase depends on intersection of unit circle and fitting line. It can area A. be obtained by calculating relative location between channel with largest volume 3. Vegetation height estimation: Based on terrain phase in Step 2, multiplied complex scattering and channel with largest surface scattering. After optimal of backscatter coherence coefficient γ with e iφ eliminates terrain phase to obtain pure volume scattering coefficient, coherence coefficient terrain γv. correction Then, result vegetation is calculated height with can be gamma calculated naught as below: and backscattering area A. ˆ sin Vegetation height estimation: Based hv on terrain phase V in Step 2, multiplied complex (25) coherence coefficient γ with e iϕ eliminates kz terrain phase to obtain pure volume scattering coherence Due to coefficient variance of γdistribution V. Then, and vegetation structure height of different can be calculated forms of vegetation, as below: accuracy of retrieval result is restricted. A diversity coefficient based on Fisher's Alpha-Diversity decomposition [31] is proposed for ĥ V = 2π ( vertical 1 2 ( structure parameter, which can be calculated k individually among different species. The alpha z π sin 1 γ V.8)) (25) factor can be used to describe relationship between Due to variance of distribution and structure of different forms of vegetation, accuracy of retrieval result is restricted. A diversity coefficient based on Fisher s Alpha-Diversity decomposition [31] is proposed for vertical structure parameter, which can be calculated individually among different species. The alpha factor can be used to describe relationship

11 ISPRS Int. J. Geo-Inf. 216, 5, of 21 between number of species and number of individuals in image scene, which indicates variance in spatial distribution and attribution of different species as in Equation (26): ( S = αln 1 + N ) α (26) where S is number of species. N is number of individuals, and α is diversity coefficient. The alpha factor can be used to represent spatial distribution and forest vegetation diversity of forest area using in-situ measurement results. Then, estimation of vegetation height is transformed into an optimization problem as below: ĥ V = 1 ( ) ( {arg ( γ V e iφ + α π 2sin 1 γ V.8))} (27) k z 4. Experiment and Discussion 4.1. Experimental Data and Description In this section, radiometric terrain correction and height estimation algorithm are tested on three data sets that include ALOS PALSAR, RADARSAT-2 and BioSAR 27 imagery. Detailed information regarding image size and location are shown in Table 1. For ALOS PALSAR quad polarization SAR data, spatial resolution is 3 m. It was acquired in April 211 in Great Xing an Mountain area in Genhe, China, and its image size is ,432 pixels. The original data set and corresponding optical image from Google earth are presented in 5a,b, respectively. RADARSAT-2 data were also acquired in quad polarization mode with a resolution of 4.7 m 5.1 m from May to August 213, in Great Xing an Mountain Yigen, China. Its image size is pixels. The main land cover is grass and crop. The average altitude of Great Xing an Mountain area is over 1 m with a slope angle from 2 to 5 degrees. This area has complex terrain characteristics of undulation and fragments; relative altitude of this area ranges from 1 to 3 m. In addition, main ground cover categories are forests and grasses with a fraction rate of nearly 75%. The dominant species are larix and white birch. BioSAR 27 PolInSAR data with P and L bands were acquired from March to April 27 with DLR E-SAR sensor over Remningstorp area in south area of Sweden. The spatial resolution is 2 m 2 m. The altitude of this forest area ranges from 12 to 145 m. Its image size is 13, pixels, and dominant species are spruce, pine and birch. The average height of vegetation in different test locations ranges from 1 m to 3 m. This area has been extensively studied and in-situ measurements were conducted, including, size of area, vegetation height, fraction, and main species, during acquisitions in 17 areas in image scene to evaluate results of different inversion methods, as shown in Table 2. Table 1. Overview of experimental SAR imagery. Sensor Test Area Center Location/Degree Latitude Longitude Image Size Pixel Resolution/m Acquisition Time ALOS PALSAR Genhe 51.6N, E 18, April 211 RADARSAT-2 Yigen 5.23N, 12.76E May August 213 BioSAR 27 Remningstorp 58.51N, 13.59E 13, April 27

12 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, of Test Test data data sets sets in in Genhe, Genhe, Great Great Xing an Xing'an Mountain Mountain area, area, China: China: ALOS ALOS PALSAR PALSAR data; data; and and optical optical data. data. Table 2. Description of in-situ measurements of forests in Remningstorp area. Table 2. Description of in-situ measurements of forests in Remningstorp area. Scene ID Size/ha Average Height/m Fraction (Number/ha) Main Species 1 Scene ID Size/ha.64 Average Height/m Fraction (Number/ha) 417 Main Species Pine Pine Pine Pine Pine Pine Pine Pine Pine Pine Spruces Spruces Spruces Spruces Spruces Spruces Spruces Spruces Spruces Spruces Spruces Birches Spruces Birches Birches Birches Birches Birches Birches Birches Birches Birches Birches For terrain 17 correction,.65 mean value, standard deviation, slope between 471 local incident Birches angle and backscattering, correction rate and calculation time are used to evaluate performance of different terrain For correction terrain correction, algorithm. mean Thevalue, correction standard rate Cdeviation, slope between local incident angle and σ is defined as decreasing percentage of backscattering, standard deviation correction between rate and original calculation SARtime imagery are used and to terrain evaluate corrected performance SAR imagery, of different as seen terrain correction algorithm. The correction rate Cσ Equation (28). S is defined as decreasing percentage of a and S b represent standard deviation of original SAR image and terrain standard correcteddeviation image, respectively. between For original vegetation SAR height imagery estimation, and terrain mean corrected valuesar and standard imagery, as deviation seen in Equation (28). Sa and Sb are also used to evaluate represent performance standard of different deviation algorithms. of The original proposed SAR method image is and runterrain corrected MATLAB image, 211a platform respectively. and polarimetric For vegetation decomposition height estimation, and interferometric mean value and are standard run in PolSARpro deviation are software. also used Comparison to evaluate methods performance including of terrain different correction algorithms. are conducted The proposed using method Gamma is run software in with MATLAB an Intel 211a Core platform and GHz polarimetric and 48. decomposition GB RAM. and interferometric are run in PolSARpro software. Comparison methods including terrain correction are conducted using Gamma software with an Intel Core GHz and 48. C σ = GB a RAM. S b 1% (28) S a Sa Sb C 1% (28) S 4.2. Experiment Results and Analysis a

13 ISPRS Int. J. Geo-Inf. 216, 5, of Experiment Results and Analysis Terrain Correction Result Table 3, s 6 and 7 presents results of different terrain correction algorithms as applied to ALOS PALSAR data for Genhe, China. To compare performance of test algorithms, scatter plots in 7 are presented to show slope between incident angle and backscatter coefficient of different terrain correction methods. In total, 1, points are selected using Gaussian random sampling method from original SAR image. Then, least squares method is investigated to obtain slope between local incident angle and backscatter coefficient. To eliminate random error of sample selection, se procedures are repeated five times, and final slope is average value. From original SAR data in 6a, we can see that mean value of backscattering is lowest, with largest standard deviation of db, which indicates that effects of terrain undulation decrease backscattering signal and increase uncertainty of backscattering estimation. The terrain undulation and backscattering coefficient is strongly correlated since backscattering value is decreasing with increase in local incident angle in original SAR imagery, and slope is For RD method, although mean value of backscattering is increasing when compared with original SAR image, standard deviation is still over 4 db; we can see some high light area in 6b. Unlike RD method, radiometric normalization approach restrains light phenomenon in SAR image; slope is.845. Thus, correction rate is increasing, but performance has a slight increase in slope area, which suggests that local terrain and backscattering are still correlated. The proposed method considers effects of terrain induced variance of incident angle, so correlation between local incident angle and backscattering diminishes. As a result, it has largest slope of.322, and correction rate on proposed method is over 3%. The computation time of RD based method, radiometric normalization method and proposed method on ALOS PALSAR data set is 23, 28 and 35 min, respectively. To investigate statistical significance of sample selection, we calculated slope using different sample points including 1, 1,, 1, and 1,, and whole points of imagery in ALOS PALSAR data with four methods. The mean slope and relative error are illustrated in Table 4. The mean slope is mean value of slope using different sample points in five instances, and relative error is calculated as in Equation (29) below, where s sample means mean value of slope using different sample points and s img represents slope calculated using whole points of image. The results in Table 4 indicate that accuracy of slope is increasing with increase in sample points. The relative error using 1, and 1,, is less than 5%, which indicates a higher statistical significance with respect to whole data set. More sample point means a greater calculation time, so we choose 1, sample points to calculate slope in our experiments. The results of proposed method on RADARSAT-2 image with four different acquisition times are illustrated in s 8 and 9 with a calculation time of 32 min on each scene. The scatter plots in 9 are obtained using same method as in 7. Table 5 gives quantative results of terrain correction on proposed algorithm. We can see that all correction rates on four images are greater than 2%, with a slope of around.6. The result for data obtained on 23 May 213 had largest slope when compared to or three images, which means that backscattering coefficients are less correlated to local incident angle. There are many factors that may cause variance in backscattering coefficients. June to August 213 was growth period for grasses and crops, which may explain why slope in 9a is larger than slope of or three results. Soil moisture is anor factor leading to variance in backscattering. The meteorological data reported that it did not rain during SAR data acquisition time, so soil moisture may not be main factor that causes variance in backscattering. This experiment indicates that vegetation parameters are also correlated to backscattering, which may also influence result of terrain correction. ( ) simg s sample error relateive = abs 1% (29) s img

14 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, of 21 (c)(c) (d) (d) 6. Terrain correction data sets setsfor forgenhe: Genhe:experiment experiment data; 6. Terrain correctionresults resultson on ALOS ALOS PALSAR PALSAR data data; RDRD based geometric correction; correction;and and(d)(d)backscatter backscatter based geometric correction;(c) (c)radiometric radiometric normalization normalization correction; optimization radiometric correction. optimization radiometric correction. 7. Cont. ISPRS Int. J. Geo-Inf. 216, 5, x; doi: FOR PEER REVIEW

15 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, of 21 (c) (d) 7. Comparison of backscattering coefficient between measurement data and experiment result 7. Comparison of backscattering coefficient between measurement data and experiment result on ALOS PALSAR data set for Genhe: experiment data; RD based geometric correction; (c) on ALOS PALSAR data set for Genhe: experiment data; RD based geometric correction; radiometric normalization correction; and (d) backscatter optimization radiometric correction. (c) radiometric normalization correction; and (d) backscatter optimization radiometric correction. Table 3. Backscatter coefficient comparison of different terrain correction method on ALOS Table 3. Backscatter coefficient comparison of different terrain correction method on ALOS PALSAR PALSAR data set in Genhe. data set in Genhe. Standard Correction Computation Method Mean/dB Slope Standard Deviation/dB Correction Rate/% Computation Time/Minute Method Mean/dB Slope Original Data Deviation/dB Rate/% Time/Minute RD Original Method Data Radiometric RDNormalization Method Radiometric Proposed Normalization Method Proposed Method Table 4. The statistical significance of sample selection. Table 4. The statistical significance of sample selection. Sample Points Method Mean Slope Relative Error/% RD Method Sample Points Method Mean Slope Relative Error/% 1 Radiometric Normalization Proposed RD Method Method Radiometric RD Normalization Method Proposed Method , Radiometric Normalization Proposed RD Method Method , Radiometric Normalization RD Method Proposed Method , Radiometric Normalization Proposed RD Method Method , Radiometric Normalization RD Method Proposed Method ,, Radiometric Normalization RD Method Proposed Method ,, Radiometric Normalization Proposed RD Method Method Whole image Radiometric Normalization.812 RD Method.153 Proposed Method.338 Whole image Radiometric Normalization.812 Proposed Method Vegetation Height Estimation Result Vegetation Since Height orbit of Estimation RADARSAT-2 Result sensor is not stable, we could not obtain stable interferometric, and this represents a lack of physical support when used for height estimation. For Since orbit of RADARSAT-2 sensor is not stable, we could not obtain stable ALOS PALSAR data set, we did not obtain in-situ measurements during acquisition time. interferometric, and this represents a lack of physical support when used for height estimation. For BioSAR 27 data, which provide in-situ measurements, we only carried out height For estimation ALOS analysis PALSAR in BioSAR data set, data we with did not P obtain and L band. in-situ Forest measurements height estimation during experiments acquisition were time. conducted For BioSAR on this 27 data data, set to demonstrate which provide performance in-situ measurements, of different methods we only and carried performance out height estimation of L and analysis P band inpolinsar BioSAR data data. with P and L band. Forest height estimation experiments were conducted on this data set to demonstrate performance of different methods and performance of L and P band PolInSAR data.

16 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, of 21 (c) (c) (d) (d) 8. Time series analysis backscattercoefficient coefficient on datadata sets sets for Yigen: Time series analysis of of backscatter onradarsat-2 RADARSAT-2 forfor Yigen: 8. Time series analysis of backscatter coefficient on RADARSAT-2 data sets Yigen: May 213; 6 June 213; (c) 1 July 213; and (d) 3 August 213. May 213; 6213; June 213; (c)213; 1 July 213; 3 August May 6 June (c) 1 Julyand 213;(d) and (d) 3 August (c) (d) 9. Time series analysis of backscatter coefficient on RADARSAT-2 data sets for Yigen: 23 May 213; 6 June 213; (c) 1 July 213; and (d) 3 August 213. (c) (d) Table 5. Time series analysis of backscatter coefficient on RADARSAT-2 data sets for Yigen. Calculation time:series 32 min. 9. Time series analysis of backscatter coefficient on RADARSAT-2 data sets forfor Yigen: Time analysis of backscatter coefficient on RADARSAT-2 data sets Yigen: May 213; 6213; June 213; (c)213; 1 July 213; (d) 3 August 213.Slope 23Date May 6 June (c)standard 1 Julyand 213; and (d) 3 August 213. Mean/dB Deviation/dB Correction Rate/% 23 May Table 65.June Time analysis of backscatter coefficient on RADARSAT-2 data23.14 sets for Yigen. 213series Calculation time: 32 min. Date 23 May June 213 Mean/dB Standard Deviation/dB Slope Correction Rate/%

17 ISPRS Int. J. Geo-Inf. 216, 5, of 21 Table 5. Time series analysis of backscatter coefficient on RADARSAT-2 data sets for Yigen. Calculation time: 32 min. Date Mean/dB Standard Deviation/dB Slope Correction Rate/% 23 May June July August shows results of vegetation height estimation for BioSAR data set. The height is divided into three intervals with an average relative height of 6 12 m, m and m based on in-situ measurement. Then, height estimation results based on phase difference, three-stage inversion and proposed method are illustrated in 1 and Table 6. The phase difference method used lowest time of.8 h, followed by three-stage inversion method with 1.1 h. However, proposed method requires more time to calculate species factor with a computation time of 1.2 h. We can see that all three approaches underestimate height, and standard deviation increases with increase in forest height. The phase difference method obtained lowest height estimation result of three methods since phase center corresponding to ground and canopy scattering mechanism was inaccurate. It is usually characterized by underestimation of canopy phase center and overestimation of ground phase center. The three-stage inversion method considers physical mechanism between ground and canopy in PolInSAR data; height estimation results improved when compared with those of phase difference method. The comparison between three-stage inversion method and proposed method indicates that variance in spatial distribution and attribution variance can affect result of height estimation. For proposed method, mean value of height estimation has lowest bias when compared with in-situ measurements, and standard deviations of three methods are nearly same. Table 6. Accuracy results of different height estimation methods on BioSAR 27 data set for Remningstorp. Method Phase Difference Three-stage Inversion Proposed method Height/m Evaluation/m 6 12 m m m Average height Standard deviation Average height Standard deviation Average height Standard deviation Calculation Time/Hour Moreover, we also compared L band and P band height estimation results using proposed method, as shown in 11 and Table 7. It is clear that bias of height estimation is increasing with increase in tree height in both P and L band data, and calculation time is 1.2 h in both images. In 6 12 m and m range, standard deviation is just around 1 m. For m, standard deviation is nearly 3 m. This may be caused by calculation error of coherence coefficient induced by penetration depth and extinction coefficient. Since penetration depth of P band is larger than that of L band, we can obtain a better estimation of terrain and canopy scattering phase center, and height estimation results for P band are better than those of L band with a higher mean value and lower standard deviation.

18 ISPRS Int. J. Geo-Inf. 216, 5, of 21 ISPRS Int. J. Geo-Inf. 216, 5, 186 ISPRS Int. J. Geo-Inf. 216, 5, of of 21 (c) (d) (c) (d) 1. Height estimation on BioSAR27 datasets for Remningstorp: Pauli RGB; phase 1. Height estimation on BioSAR27 datasets for Remningstorp: Pauli RGB; phase difference; (c) three-stage inversion; and (d) species diversity based inversion. difference; (c) (c) three-stage inversion; and (d) (d) species diversity based inversion. 11. Height of proposed method on BioSAR 27 with P and L band for Remningstorp: Height Height estimation P estimationof proposed method on BioSAR 27 with and band for band; and of L band. proposed method on BioSAR 27 with P and L band for Remningstorp: P band; and and L band. band.

19 ISPRS Int. J. Geo-Inf. 216, 5, of 21 Table 7. Accuracy results of proposed method with P and L band BioSAR data set for Remningstorp. Band P L Evaluation/m Height/m 6 12 m m m Average height Standard deviation Average height Standard deviation Calculation Time/Hour Conclusions This paper presents a radiometric terrain correction method and forest vertical structure parameter estimation approach in PolInSAR imagery. Terrain undulation is considered, and a form of scattering mechanism classification and a scattering center optimization method have been proposed to improve accuracy of backscattering coefficient estimation. The purpose is to eliminate influence of complex terrain. In forest height estimation procedure, species diversity factor is exploited in inversion model to describe spatial distribution as well as scattering variance of different species. To validate effectiveness of scattering center optimization approach for terrain correction, a Geometric-based method and radiometric normalization method have been applied to on ALOS PALSAR and RADARSAT-2 data sets. The result shows that proposed method obtains highest mean value and lowest standard deviation of backscattering coefficient, and correction rate of proposed method is also highest. In tree height estimation experiment, proposed method obtain get most accurate result for tree height at different height intervals, and P band inversion result outperforms L band inversion result with higher height estimation and lower standard deviation. The main contributions of proposed algorithm include two aspects. A scattering mechanism classification and scattering center optimization approach is proposed to eliminate influence of terrain undulation. The spatial distribution and scattering variance of different species have been considered in height inversion procedure to improve accuracy of height estimation. Furr work will be extended to analyze effect of surface soil moisture on backscattering and carbon cycling will be modeled in forest area, which has been hindered by terrain undulation. A robust and operational model will be constructed for analyzing relationship between biomass estimation and carbon cycling in forest area. Acknowledgments: The work was supported by NSFC grant (No , and No ) and National Key Basic Research and Development Program of China (973 Program) (No. 213CB73344). The data sets are provided by Nation Basic Research Programs of China (973 Program) Dynamic analysis and modeling of remote sensing information for complex land surface (213CB7334). We also thank ESA for providing BioSAR data sets and corresponding in-situ measurements data via Dragon Program. Author Contributions: Yu Zhang, Chu He and Xin Xu conceived and designed experiments; Yu Zhang performed experiments; Yu Zhang and Dong Chen analyzed data; and Yu Zhang wrote paper. Conflicts of Interest: The authors declare no conflict of interest. Abbreviations The following abbreviations are used in this manuscript: SAR PolInSAR ALOS PALSAR RADARSAT BioSAR RVoG PDF RD GEC Syntic Aperture Radar Polarimetric Interferometry Syntic Aperture Radar Advanced Land Observing Satellite Phased Array type L-band Syntic Aperture Radar Radio Direction and Range Satellite Biomass Syntic Aperture Radar Random Volume-over-Ground Probability Density Function Range Dopple Geocoding Ellipsoid Corrected

20 ISPRS Int. J. Geo-Inf. 216, 5, of 21 GTC GCP DEM RMSE LUT Geocoded Terrain Corrected Ground Control Points Digital Elevation Model Root Mean Squared Error Looked Up Table References 1. Yang, X.; Huang, J.; Wang, J.; Wang, X.; Liu, Z. Estimation of vegetation biophysical parameters by remote sensing using radial basis function neural network. J. Zhejiang Univ. Sci. A 27, 8, [CrossRef] 2. Roudgarmi, P.; Monavari, M.; Feghhi, J.; Nouri, J.; Khorasani, N. Environmental impact prediction using remote sensing images. J. Zhejiang Univ. Sci. A 28, 9, [CrossRef] 3. Dixon, R.; Solomon, A.; Brown, S.; Houghton, R.; Trexier, M.; Wisniewski, J. Carbon pools and flux of global forest ecosystems. Science 1994, 263, [CrossRef] [PubMed] 4. Cloude, S.; Papathanassiou, K. Polarimetric SAR interferometry. IEEE Trans. Geosci. Remote Sens. 1998, 36, [CrossRef] 5. Gabriel, J.; Schmullius, C.; Papathanassiou, K.; Darizhapov, D.; Tathkov, G.; Tsybjitov, T.; Boerner, W. Tree height extraction using Polarimetric SAR interferometry. In Proceedings of IEEE 1999 International Geoscience and Remote Sensing Symposium (IGARSS 99 Proceedings), Hamburg, Germany, 28 June 2 July Yamada, H.; Yamaguchi, Y.; Rodriguez, E.; Kim, Y.; Boerner, W. Polarimetric SAR Interferometry for forest canopy analysis by using super-resolution method. In Proceedings of IEEE 21 International Geoscience and Remote Sensing Symposium (IGARSS 1), Sydney, Australia, 9 13 July Wang, Y.; Chen, H.; Peng, Y.; Wan, Q. Theory and Algorithm of Spatial Spectrum Estimation; Qinghua University Press: Beijing, China, Guillaso, S.; Gsrestier, F. Deformation estimation on low coherence areas by means of Polarimetric differential SAR interferometry. In Proceedings of 215 8th International Workshop on Analysis of Multitemporal Remote Sensing Images (Multi-Temp), Annecy, France, July Oveisgharan, S.; Saatchi, S.; Hensley, S. Sensitivity of Pol-InSAR measurements to vegetation parameters. IEEE Trans. Geosci. Remote Sens. 215, 53, [CrossRef] 1. Papathanassiou, K.; Hajnsek, I.; Moreira, A.; Cloude, S. Interferometric SAR Polarimetry using a passive polarimetric microsatellite concept. In Proceedings of 22 IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2), Toronto, ON, Canada, June Xin, L. Information Extraction Methods and Application Study of Polarimetric SAR Interferometry. Ph.D. Thesis, Institute of Remote Sensing and Digital Earth Chinese Academy of Sciences, Beijing, China, 22. (In Chinese) 12. Zou, B.; Zhang, L.; Wang, W.; Sun, D. Forest parameters inversion using Polinsar data based on genetic algorithm. In Proceedings of 26 IEEE International Geoscience and Remote Sensing Symposium, Denver, CO, USA, 31 July 4 August Angiuli, E.; Frate, F.; Vecchia, A.; Lavalle, M. Inversion algorithms comparison using L-Band simulated 462 Polarimetric interferometric data for forest parameters estimation. In Proceedings of 27 IEEE International Geoscience and Remote Sensing Symposium, Barcelona, Spain, July Cloude, S.; Papathanassiou, K. Three-stage Inversion Process for Polarimetric SAR Interferometry. IEE Proc. Radar Sonar Navig. 23, 15, [CrossRef] 15. Cloud, S. PolInSAR regularization using dual frequency interferometry. In Proceedings of International Radar Conference, Adelaide, Australia, 3 5 September Shao, Z.; Yang, N.; Xiao, X.; Zhang, L.; Peng, Z. A multi-view dense point cloud generation algorithm based on low-altitude remote sensing images. Remote Sens. 216, 8. [CrossRef] 17. Lebrl, F. Modeing of single image geometric problems. In Radargrammetric Image Processing; Artech House: London, UK, 199; pp Schreier, G.; Meier, E. SAR data acquisition and image formation. In SAR Geocoding: Data and Systems; Wichmann: Osnabrück, Germany, 1993; pp

21 ISPRS Int. J. Geo-Inf. 216, 5, of Meier, E.; Graf, C.; Nuesch, D. Generation of geocoded spaceborne SAR image products. In Proceedings of 12th Canadian Symposiumon Remote Sensing, International Geoscience and Remote Sensing Symposium, (IGARSS 89), Vancouver, BC, Canada, 1 14 July Hoekman, D.; Reiche, J. Multi-model radiometric slope correction of SAR images of complex terrain using a two-stage semi-empirical approach. Remote Sens. Environ. 215, 156, [CrossRef] 21. Sekhar, K.; Senthil, K.; Dadhwal, V. Geocoding RISAT-1 MRS images using bias-compensated RPC models. Int. J. Remote Sens. 214, 35, [CrossRef] 22. Bayer, T.; Winter, R.; Schreier, G. Terrain influences in SAR backscatter and attempts to ir correction. IEEE Trans. Geosci. Remote Sens. 1991, 29, [CrossRef] 23. Small, D.; Holecz, F.; Meier, E.; Nüesch, D.; Barmettler, A. Geometric and radiometric calibration of RADARSAT images. In Proceedings of Geomatics in Era of RADARSAT, Ottawa, ON, Canada, 24 3 May Small, D.; Holecz, F.; Meier, E.; Nuesch, D. Absolute radiometric correction in rugged terrain: A plea for integrated radar brightness. In Proceedings of 1998 IEEE International Geoscience and Remote Sensing Symposium (IGARSS 98), Seattle, WA, USA, 6 1 July Frulla, L.; Milovich, J.; Karszenbaum, H.; Gagliardini, D. Radiometric corrections and calibration of SAR images. In Proceedings of 1998 IEEE International Geoscience and Remote Sensing Symposium (IGARSS 98), Seattle, WA, USA, 6 1 July Carlos, L.; Papathanassiou, K. Cancellation of scattering mechanisms in POLinSAR: Application to underlying topography estimation. IEEE Trans. Geosci. Remote Sens. 213, 51, He, C.; Li, S.; Liao, Z.; Liao, M. Texture classification of PolSAR data based on sparse coding of wavelet polarization textons. IEEE Trans. Geosci. Remote Sens. 213, 51, [CrossRef] 28. Minh, D.; Toan, T.; Rocca, F.; Tebaldini, S.; d Alessandro, M.; Villard, L. Relating P-band syntic aperture radar tomography to tropical forest biomass. IEEE Trans. Geosci. Remote Sens. 214, 52, [CrossRef] 29. Schreier, G.; Meier, E. Precise terrain corrected geocoded images. In SAR Geocoding: Data and Systems; Wichmann: Osnabrück, Germany, 1993; pp Ballester-Berman, J.D.; Lopez-Sanchez, J.M. Applying freeman-durden decomposition concept to polarimetric SAR interferometry. IEEE Trans Geosci. Remote Sens. 21, 48, [CrossRef] 31. Saatch, S.; Buerrmann, W.; Steege, H.; Mori, S.; Smith, T. Modeling distribution of Amazonian tree species and diversity using remote sensing measurements. Remote Sens. Environ. 28, 112, [CrossRef] 216 by authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under terms and conditions of Creative Commons Attribution (CC-BY) license (